Episode 2

full
Published on:

4th Jun 2026

Why Digital Twins Need an Intelligence Layer

Michael Jansen and Dr. Prasanta Bose join Evan Troxel and Randall Stevens to talk about what it takes to put an intelligence layer on top of a digital twin. Bose traces the idea back through reinforcement learning, Lockheed Martin satellites, and Starbucks before explaining why a real twin needs both a sensing layer and a cognitive one. They get specific about the decisions behind TwinMaster: refusing to build another design authoring tool and instead embedding their Arch-e copilot inside Revit, Archicad, and MicroStation; building a semantic, systems-oriented model so the AI can reason about a wall as more than two planes; and tuning existing models with context rather than training their own.

This episode is especially relevant for design technologists, BIM leads, and AEC software teams weighing how AI actually fits into established tools instead of replacing them. Jansen makes the case that architects, who create the original twin, could sell and maintain it as an ongoing service and move past one-time fixed fees. You will come away rethinking where the value sits after the drawings are done.

Episode Links:

Connect with the guests

  • Dr. Prasanta Bose, CEO and co-founder — LinkedIn
  • Michael Jansen, Chief Business Officer and co-founder — LinkedIn

TwinMaster

Tools and ecosystems mentioned

People and ideas referenced

Watch this episode on YouTube.

-----

The Confluence podcast is a collaboration between TRXL and AVAIL, and is produced by TRXL Media.

Transcript
Randall Stevens:

Welcome to another episode of the Confluence podcast.

2

:

I'm Randall Stevens, and of course

I've got, uh, my sidekick Evan Troxel.

3

:

we're happy to have Michael Jansen and Dr.

4

:

Prasanta Bose from,

uh, Twinmaster with us.

5

:

So we're gonna dig in, uh,

uh, to everything that they've

6

:

been doing with Twinmaster.

7

:

Welcome guys.

8

:

Michael Jansen: Thank you.

9

:

We, it's lovely to be here.

10

:

We

11

:

appreciate the invitation.

12

:

Randall Stevens: let's kick this off.

13

:

You know, Evan gave the kind of,

uh, background of you two, Sue.

14

:

Um.

15

:

You know, I'm assuming Prasanta that

a lot of, a lot of what has gone into

16

:

the thinking that became Twinmaster

originated with your work when you

17

:

were doing work at Starbucks or maybe

even all the way back at Lockheed.

18

:

But maybe give us that, let's start

there with kind of the background.

19

:

What led to the development of Twinmaster?

20

:

What, what

21

:

was the IMP impetus for that?

22

:

Dr. Prasanta Bose:

23

:

Yeah.

24

:

I think, you know, it dates

back, maybe even earlier.

25

:

Uh, basically it dates

back from my graduate days.

26

:

Working with, uh, people

like Andy Barto Right.

27

:

Is a cheering award.

28

:

Yeah.

29

:

And, uh, rich Sutton, you know,

reinforcement learning is all

30

:

about intelligent systems.

31

:

Uh, so that thinking matured

or you can say took more.

32

:

You can, uh, uh, center in front in, in

Lockheed Martin, uh, at Lockheed Martin,

33

:

uh, uh, the, if you look at it, it's

a systems world of space systems and

34

:

aircraft systems and, and, and my, with

my interest in making them intelligent,

35

:

right, uh, led me into, you can say this

overall different areas, but primarily

36

:

to do with how do you make it autonomous?

37

:

Uh, you know, the cost

is a big driver, right?

38

:

Uh, so, so when you're looking at,

you know, instead of having one

39

:

satellite per mission, how do you

make it a reconfigurable satellite?

40

:

So, so you're looking at kind

of a self, uh, configuring.

41

:

Kind of a, a payload that's,

you know, one concept, right?

42

:

All the way to how do you look at

formation flying of, of a sort of drones.

43

:

In this case, it was

satellite kill, right?

44

:

Because we, I was playing a major

role in this thing called MKV,

45

:

multiple kill vehicles, right?

46

:

How do you define a kill basket and

how this, these guys self-organize

47

:

depending on, you know, what's the

target, what's the flare, right?

48

:

And go after the target because, you know,

so there are these very hard problems of

49

:

control coming in, intelligence coming in.

50

:

How do you share the information?

51

:

How do you self-organize?

52

:

So that is one.

53

:

And then the other part was I was

simultaneously doing a ton of, you

54

:

know, you can say, and NASA projects.

55

:

And nasa basically this thing

was part of the earth science.

56

:

Looking at, you know, monitoring the suns.

57

:

Weather, climate, you know, the flares,

how that influences, you know, the, the

58

:

earth's weather, uh, climate and weather.

59

:

As well as kind of thinking about how you

monitor forest fires out of that born the,

60

:

in the notion of digital twins, right?

61

:

The reason was, this is where I was

thinking that as if the satellite was

62

:

sensing the data, if I can make it

a kind of a virtual twin, which is

63

:

addressable, uh, you know, giving me,

you know, queryable questionable, right?

64

:

Uh, anything, the task I give as

if you got, you got a hardware,

65

:

you know, the satellite is watching

over something as well as the data

66

:

that is giving you the intelligence.

67

:

And then being able to coordinate,

you know, let's say something

68

:

where lightning is happening.

69

:

Somebody is speaking up some

data about, you know, what's the

70

:

dryness of this place, right?

71

:

So this was, gave rise to this, you

know, this, uh, you can say this, uh,

72

:

digital twin cons construct, right?

73

:

And this was early way back,

right when I first start literally

74

:

maybe two or three years into

Lockheed, uh, close to:

75

:

Uh, it went on to develop into, into

looking at Locke internally as we

76

:

are, you know, creating these, uh,

you can say very complex systems.

77

:

Where people are basically keeping

things in Excel spreadsheets

78

:

and complex MATLAB models.

79

:

And so there was no way to kind

of connect in a seamless manner.

80

:

So I thought digital twin

is kind of the DNA, right?

81

:

If, if I can keep, keep the kind of the

intelligence there and as people are

82

:

working with it, is giving you feedback.

83

:

So there are these, uh, things that

develop from there onto, you know, as

84

:

I started applying into these DARPA

things of self-organized robots,

85

:

right, called OIDs, right, where

they would, you know, you set them

86

:

in a building and they self configure

to, you know, maintain a network.

87

:

You know, if somebody dies,

they take over, right?

88

:

So if you look at it, these are,

you can say at a macro level of,

89

:

you know, things that was seeded way

back in when I was doing at UMass,

90

:

working with people like Mark Rwe.

91

:

Uh, then there was another

person looking at, you know, the

92

:

cybernetic model of the brain.

93

:

Okay.

94

:

How does spatial reasoning, how do you,

you know, I was even looking at, I mean,

95

:

I was taking courses in neurobiology of

learning of how does a, basically a rat,

96

:

you know, runs a maze kind of thing.

97

:

Right?

98

:

What gets wire, what's reinforcement

learning doing to it, right.

99

:

And was my thesis was

in that kind of thing.

100

:

But anyway, uh, so these constructs

at Lockheed, uh, uh, I mean

101

:

it's primarily to make scalable.

102

:

Autonomous systems, which are

self-organizing and can actually

103

:

work independently, you know,

with obviously human in the loop.

104

:

Right?

105

:

Uh, there was other kinds of things

that I did while that Lockheed, uh,

106

:

you know, uh, like very large scale

sensor networks, you know, this small

107

:

mulch they used to call came out of

the Berkeley, you know, uh, thing.

108

:

Very tiny devices.

109

:

You spin them in thousands

and they self configure.

110

:

Right.

111

:

Out of that came different, you know,

I mean, I was a lot of doing a lot of,

112

:

I would say very creative thinking.

113

:

I would say, you know what I mean?

114

:

Very entrepreneurial thinking.

115

:

Okay.

116

:

Uh, and, and, uh, because of that,

I also thought maybe, you know what,

117

:

every time I was doing this proposal,

and it's like a crile to grave kind of

118

:

thing, you, you come up the construct,

go for the, you know, BNP money, do

119

:

the project, build in a very intense

competitive environment, right.

120

:

And then go on to, you know,

deliver on it in literally, you

121

:

know, like a field demonstration.

122

:

They said, you know

what, let's do a startup.

123

:

Get out and do take one of

these ideas into startup.

124

:

That's where I basically

co-founded falconry.

125

:

It is still active.

126

:

I'm also part of it.

127

:

Right.

128

:

Uh.

129

:

And from there I all went on to Starbucks.

130

:

Starbucks is a, you know, again,

uh, looking, uh, from an angle, is

131

:

that because my training had been

into looking at these different

132

:

disciplines of the systems world, right?

133

:

I, I forgot to tell you in the, at

Lockheed also, uh, I had the, you

134

:

know, fortune of basically getting

this Lockheed white, you can say

135

:

one person being selected to go to

Santa Fe Complex Systems Institute.

136

:

Again, looking at the big

picture of the world, right?

137

:

As a complex systems as a network

of things like the brain think

138

:

as if working at scale, right?

139

:

I learned a lot about, you know, things.

140

:

So, so, so same thing when I went into

Starbucks saying, Hey, you know what?

141

:

This is, you know, thousands

of stores worldwide.

142

:

Sending data, you know, giving, delivering

to coffee is a wonderful macro brain.

143

:

If I care the brain of Starbucks, and

you know, I said, Hey, you know what?

144

:

Every coffee that goes

out has the right quality.

145

:

I have visibility.

146

:

What a store is doing.

147

:

It's like the coffee to have

digital, you know, IO you know,

148

:

kind of a twin of the store and

then a macro, you know what I mean?

149

:

Store, right?

150

:

Uh, uh, that gives you

visibility, you know, which,

151

:

which store is doing well, right?

152

:

And, and, and so on, right?

153

:

Uh, uh, from there, uh, Starbucks,

I went to equity space, right?

154

:

Equity is a, is building is lights.

155

:

You know, you've heard of, uh, all the

big airports and everything, right?

156

:

Uh, so from there, I mean, at

ati I was doing two things.

157

:

You know, it's like this thing,

uh, to do with how do you, uh.

158

:

How do you price, what's, what's a,

what's a competition on pricing engine?

159

:

That's one.

160

:

The other one was again, uh, making the

lights themselves intelligent so that

161

:

consider basically Walgreen and you are

moving around, you know, they can tell,

162

:

you can collect intelligence, which

part of the aisle is getting heavily

163

:

visited so that you can reorganize

and make easy for the customer.

164

:

Right?

165

:

So the underpinning of all of this

is this notion of creating Twin

166

:

as a way to model the environment

and the, it has a life of its own.

167

:

So it was a structure and, and

obviously intelligence behind it.

168

:

And out of this Born, you know, I met,

uh, Michael, uh, to create Twinmaster.

169

:

And, you know, it's a brilliant, because

it's a, it's, I would say, convergence

170

:

of multiple disciplinary ideas.

171

:

And there's a core theme, you

know, going through that thing of

172

:

how do you create smart systems,

intelligent things so that you make

173

:

the, you know, a little life easier.

174

:

Randall Stevens: I guess my first

question I'd like to dig in on is, um,

175

:

you know, for me it's easy to think

about the digital twin being a mirror

176

:

of, of, of a state, of something.

177

:

And then data, IOT data or

whatever can be attached to that.

178

:

Uh.

179

:

So I get that part.

180

:

Explain a little bit more about how

this idea of, like, that the systems

181

:

are reconfiguring or is that just,

uh, the idea that, that it's, it is

182

:

alive and the state is changing all the

183

:

time.

184

:

Can you

185

:

Dr. Prasanta Bose:

186

:

Yeah, so, right.

187

:

so the view is like this, right?

188

:

I, I always look view for the world

from a control theory standpoint.

189

:

That's also we do, right?

190

:

It's like a, basically I

call as the plant layer.

191

:

Plant is basically just like the robot,

you know, actuating kind of thing, right?

192

:

Or, or, uh, or a or a, even a

manufacturing plant producing stuff.

193

:

And then there's a cognitive layer, right?

194

:

Or you can see the

intelligence layer, right?

195

:

So there's a feedback going on.

196

:

So think about this IOT or you know,

or twin that you're building has

197

:

both these aspects in that there

is the part which is a sensory

198

:

layer giving you the information.

199

:

That is coming in, streaming in, right?

200

:

You're doing analytics, you're

forming a world model of that thing.

201

:

Because without model, you

cannot do model predictive thing.

202

:

Because at the end of the day, we

right from a form, a constructive

203

:

model of that world or space, right?

204

:

So that we can work, you know,

we can act in the world, uh, in

205

:

the world in a proper manner.

206

:

So.

207

:

When you look at the twin, you've

got both sides of the story, right?

208

:

The intelligence layer and the

sensing and actuation layer, right?

209

:

In order to make it a real in twin,

because at the end of the day you are, you

210

:

are making the digital twin to empower.

211

:

You are now creating a system, just like

we, if you take the brain off, right?

212

:

We just like the, you got the substrate

or the physical substrate, right?

213

:

So now with the digital twin,

the digital layer, right?

214

:

Which has this both the sensing

part keep in mind, right?

215

:

Because that's an important

thing, and the actuation, right?

216

:

And that intelligence, which is basically

making sense of the data that's pouring

217

:

in from the sensory data, forming

a model and then closing the loop.

218

:

Hey,

219

:

what should I

220

:

Randall Stevens: Yeah, I was, I was,

uh, you know, when you were first

221

:

describing that, I guess my, uh,

you know, even what you were talking

222

:

about doing with Lockheed, it's like.

223

:

One of the things I think that, that,

at least I think about with a digital

224

:

twin is it lets me run simulations.

225

:

It lets me do things.

226

:

But, but really that cognitive

layer is just an advanced

227

:

form of that.

228

:

Like, you know,

229

:

Dr. Prasanta Bose:

230

:

Exactly.

231

:

Exactly.

232

:

Exactly.

233

:

I mean, that's why I took that,

you know, moved that forward

234

:

because I was motivated that.

235

:

If I, it is like, it is, like it is

a, it is basically, again, going back

236

:

to the central idea that in order for

me to act in the world, I need to have

237

:

a model of the world, the physics.

238

:

So you, you, you have seen for example,

uh, the, this theory now going in,

239

:

where is the limit of ai, right.

240

:

Kinda stuff.

241

:

Does it have a stateful model?

242

:

Right.

243

:

The reason is, is this what they call

even the, in the in controlled thetic

244

:

world called model predictive control.

245

:

That is the model is predicting

what is going to happen to the

246

:

state, the history, you know,

based on the past history, right?

247

:

And, and they, there are different

constructs in that it comes out

248

:

of deep control theory, you know,

in, in aircraft everywhere, right?

249

:

This, this receding frontier.

250

:

The frontier is moving because

the world is dynamically changing.

251

:

But your mall is tracking that and

saying, Hey, wait a minute, you know,

252

:

the next few this thing happens, thing

this is gonna then I, I, I'm prepared.

253

:

I'm not react.

254

:

You know what I mean?

255

:

Okay.

256

:

Sorry, I I keep

257

:

Randall Stevens: No, this is cool.

258

:

This is the fun stuff

259

:

Dr. Prasanta Bose:

260

:

if you get

261

:

Randall Stevens: This

262

:

Dr. Prasanta Bose:

263

:

I so many directions you have to

264

:

Randall Stevens: you you've obviously

been, you know, you've been working

265

:

on this 25 plus years and have a

level of understanding about it.

266

:

I'll just say from an AC,

practical, but academic as well.

267

:

It's hard for some of us to put

language to it in the right way.

268

:

We kind of sense it, but

269

:

it's hard to, uh, so anyway,

thanks for helping to explain that.

270

:

Dr. Prasanta Bose:

271

:

No, no.

272

:

That, but that, that, but

273

:

what you, what you, what you're asking

is also because I also wrestle, right?

274

:

Uh, I, I have found, because I've

studied, you know, I used to study

275

:

a lot of philosophy like, you

know, uh, women fire and dangerous

276

:

thing, but George Lakoff, right?

277

:

You know, how does you

know semantics grow, right?

278

:

How does, you know, if you

look at that neuron, you know,

279

:

anchoring that is going on, right?

280

:

Uh, because I, as I was doing deeply

in, when I was doing my PhD, I

281

:

was vacating between, you know, I

mean, should I go into this area

282

:

of learning and neuro computing, or

should I be living in the logic layer?

283

:

You know, there's this

logics of reasoning, right?

284

:

So anyway, but, but, uh, you know, you,

we, I, I had to wrestle and get a visceral

285

:

understanding, you know, what I call

as, can I visualize how is this working?

286

:

You know what I mean?

287

:

And I can explain to myself

otherwise, you know, I cannot build

288

:

anything.

289

:

Randall Stevens: Right.

290

:

You gotta have a mental model Right.

291

:

To, to

292

:

Dr. Prasanta Bose:

293

:

Exactly what you just said.

294

:

I keep on asking these guys, do you have

a mental model, what you're programming?

295

:

Do you understand what you know?

296

:

If you're talking about an AI

agent, do you have a mental

297

:

model, how it works in the world?

298

:

How are you gonna, if you don't,

you'll be, you won't have any freak

299

:

idea to how to create something

300

:

new out

301

:

Randall Stevens: I don't know if y'all

heard, uh, I was watching one of the,

302

:

uh, all hands meetings last week that

they, that they broadcast from XAI,

303

:

but they were talking about the, the

advanced in their model for coding model.

304

:

And then after they talked

about that, Musk said.

305

:

He thinks by the end of the year that

they're gonna be skipping the code

306

:

and basically producing a binary,

like the AI will create a binary.

307

:

There is no code, so it's like, it's

just hard to wrap your head around,

308

:

like, you know, it's basically a

neural net is gonna have an input,

309

:

an output, and you have to be able to

310

:

test against that.

311

:

Anyway, it's just interesting.

312

:

Dr. Prasanta Bose:

313

:

No, no, no.

314

:

I mean,

315

:

uh uh, it's, I mean, at the end of the

day, right, at the end of the day, if

316

:

you look at it, right, fundamentally

what has happened is that, you know,

317

:

that expressive world of language, image

and everything, right, got tokenized.

318

:

That means vectorized, right?

319

:

Okay.

320

:

You have heard those things, right?

321

:

You know, and being embedded, right?

322

:

And through this encoder decoder

architecture, it is generating the, you

323

:

know, the a text, which is consistent

with the context and what you get, right?

324

:

I mean, you know, I'm not getting into

the mechanics of how that happens,

325

:

but that's the reality, right?

326

:

It got

327

:

codified.

328

:

it got compiled into,

329

:

Randall Stevens: that's in

technology, you know, good, good

330

:

technology and good software should

be indistinguishable from magic.

331

:

It is like magic.

332

:

You're feeding this stuff in

333

:

and it's like al and, uh, anyway.

334

:

Ev Evan, Evan, you had, uh, comments or

335

:

Evan Troxel: maybe Michael can chime in

and just give us kind of the vision of

336

:

Twinmaster to set a little bit of context

337

:

Michael Jansen: you know, Twinmaster

was conceived as, uh, a company

338

:

that could figure out how to

apply AC uh, AI in the AEC space.

339

:

Presanta didn't talk a lot about his

AI background, but he has two PhDs in,

340

:

in ai and it goes back to the eighties.

341

:

He's been at it for a very long

time and he's led initiatives in

342

:

AI for, I guess 35 or 40 years now.

343

:

40 years.

344

:

It's been a long time.

345

:

My previous company was a company

called City Zenith that was dedicated

346

:

to trying to apply digital twin

technology at a city scale to help

347

:

cities to become more efficient.

348

:

And one of the things that we

struggled with and didn't have

349

:

at the time we were building that

company was something like ai.

350

:

So by the time I met PTO, we said,

you know, we really need to rethink

351

:

everything that we're doing here

'cause AI is gonna change everything.

352

:

So how do we, how do we develop a

product that can leapfrog these existing.

353

:

Design offering tools that are largely

algorithm and deterministic and into

354

:

this world of inferential AI that we

really think is gonna be the future.

355

:

So Twinmastery was born as an idea that we

could create this type of copilot, that we

356

:

could embed into existing modeling tools

that architects and engineers already love

357

:

to basically amplify them and make them

smarter by allowing the co-pilot to do all

358

:

of the, the complex, what we call multi

objective reasoning or, or the thinking.

359

:

So today, a lot of, um, what you see in

some of these tools, uh, they will have

360

:

plugins that will, for example, uh, um,

optimize an energy model or a carbon

361

:

model or um, a constructability model.

362

:

But you know, the way that architects

and engineers think is really

363

:

all these KPIs at the same time.

364

:

So you have to develop a tool that

can think across these different.

365

:

Performance metrics and then reason on

top of them to be able to give you the

366

:

ultimate, the kind of optimal output.

367

:

So it might then need to be

able to consider carbon and cost

368

:

and energy and maybe a comfort

condition all at the same time.

369

:

So how do you do that?

370

:

That was the thesis behind Twinmaster.

371

:

We didn't wanna replace the

existing modeling tools.

372

:

We think there's already too many

modeling tools in the market as it is,

373

:

and they have very large, entrenched

audiences of millions of existing users.

374

:

So in the beginning of the company,

we thought maybe we would just try

375

:

to, you know, sort, uh, elbow our

way into that market, um, by seeing

376

:

if we could develop an independent

product that would, um, eventually

377

:

become a kind of, uh, effective,

another kind of design authoring tool.

378

:

But as we got deeper into it, we

realized that we were better off,

379

:

uh, um, uh, sort of amplifying

and supercharging tools like.

380

:

Micro Station 3D and ArchiCAD and Revit

and these other tools already out there.

381

:

So ultimately that's how we

began positioning ourselves

382

:

Randall Stevens: uh, was that?

383

:

A forced decision or just

a kind of a, a business,

384

:

uh, business

385

:

model decision, right.

386

:

To build your own ing tool.

387

:

Dr. Prasanta Bose:

388

:

maybe, yeah.

389

:

Michael Jansen: me answer

that because it's, it's an

390

:

Dr. Prasanta Bose:

391

:

okay.

392

:

Yeah, go ahead.

393

:

Michael Jansen: question.

394

:

It's, it's a fundamental question.

395

:

It's a fundamental

396

:

question.

397

:

Um, and very well asked.

398

:

Um, we really didn't see a business

in starting another design authoring

399

:

tool.

400

:

Randall Stevens: because

there's several, right?

401

:

That are That are on,

402

:

Michael Jansen: There's so many.

403

:

We, we, we need not name them.

404

:

There are so many BIM 1.0,

405

:

BIM 2.0,

406

:

multiple tools.

407

:

We actually did a hard look, uh, and

hired a, um, a third party consultant

408

:

out of New York that to study the marker

for us and, and to, to report back to

409

:

us what was the actual uPresantake.

410

:

In the market of all these tools from

the big ones to the newer ones, right.

411

:

And what we found was that the BIM 2.0

412

:

tools just were not getting

uPresantake and still, and we

413

:

felt that, we kind of asked a hard

question, why is that the case?

414

:

And then we started talking to CIOs

and CTOs around the market and they

415

:

would tell us, you know, we just

can't afford to rip out tools that

416

:

we've already relied upon for years.

417

:

There's, there are well established

ecosystems around these tools

418

:

already training protocols.

419

:

Uh, it'd be very difficult to

remove these tools and they are

420

:

improving incrementally as it is.

421

:

So it's not as over unhappy, even though

perhaps some of them could be better.

422

:

But what we felt was missing was in,

uh, trying to apply intelligence to the

423

:

way that these tools were being used.

424

:

So we felt maybe it's a little bit

smarter to partner with them than to

425

:

compete against them, that it was just

a fundamental decision that we made.

426

:

So in the, in the beginning of

the company, we were terrified.

427

:

To go up and talk to people

at Autodesk in Bentley.

428

:

And you mentioned earlier, uh, McNeil,

uh, Trimble Nemechek chaos, some of

429

:

these companies, because we didn't

know, you know, we're a small company,

430

:

we didn't know if we had anything to

offer, to be honest, and if we could

431

:

even be considered as a potential, you

know, another player in their universe.

432

:

And so we very gradually, we started,

uh, um, socializing in this concept

433

:

in the market about a year ago.

434

:

And then we stopped hiding from everybody

and actually participated in a few shows.

435

:

And we took part in the a i A

show in Boston and, uh, about

436

:

eight months ago or so, and we

won best innovation of that show.

437

:

And what, what was, what surprised

me is that three of the big six AEC

438

:

software companies came over to meet us.

439

:

I mean, at a senior level, there was

clearly curiosity and we realized we

440

:

actually had something in hand with

our approach that was unique and, and.

441

:

Uh, significantly far enough ahead

in the market that they would rather

442

:

partner with us than try to rebuild it.

443

:

I think that was the beginning of this

new model that we adopted, which was

444

:

let's partner, let's embellish existing

tools that already are in the marketplace.

445

:

So since then, that's been

the path that we're on.

446

:

I think, you know, some of you know

that last week Bentley announced a

447

:

global partnership with us publicly.

448

:

There will be more, I can't announce

them all right now, but we think there's

449

:

a very strong chance that we'll end

up working with most of the major AEC

450

:

software players by the end of the year.

451

:

And what we're doing is we're

developing an integration with them

452

:

first, which allows us to connect

to their ecosystems of tools.

453

:

And typically they have some type

of, um, data hub or data lake type

454

:

product, like a Trimble connector, an

A CC or an I twin then, and then they

455

:

have certain popular modeling tools.

456

:

So again, we did research to

see which of these modeling

457

:

tools are popular they used.

458

:

And we looked at each company and said,

okay, they've got this, this, and this.

459

:

And we started to focus on those first.

460

:

We became an Autodesk technology partner

in November of last year, which was

461

:

the first formal, um, I think, uh,

arrangement that we made with any of them.

462

:

And now it's kind of, you know, I can't

name names yet 'cause it might be a little

463

:

bit ahead of where some of the marketing

teams are at, but there's a strong chance

464

:

that we'll be working with everybody

before you know it, you know, and so that

465

:

allows us to focus on just making Archie

our co-pilot as good a tool as it can be,

466

:

rather than trying to recreate the, the,

these design authoring tools that are

467

:

already fairly successful in the market.

468

:

And that's, that's where we're focusing.

469

:

Randall Stevens: Do, do you end up, uh,

pulling that data geometry, but also

470

:

metadata, like into your own neutral

471

:

database or place to do the processing?

472

:

Dr. Prasanta Bose:

473

:

so, uh, if you, if you basically,

uh, the, the key underlying

474

:

principle by which AI works right,

is this aspect of semantics, right?

475

:

Uh, the generalization or specialization,

how the kq mechanisms work under the

476

:

hood of an encode decoder, right?

477

:

So semantics is important, right?

478

:

So therefore, if I, if, if I

want to create a general purpose

479

:

engine like Archie kind of thing,

I need to create a kind of an,

480

:

uh, you can say that annotated,

semantically, annotated model, right?

481

:

So that it becomes a common, common

layer, what you just said, that that

482

:

means, because, you know, I can get from

483

:

Randall Stevens: Multiple sources.

484

:

Yeah.

485

:

Dr. Prasanta Bose:

486

:

I can get from auto death, right?

487

:

Uh, and the other thing, this is the

most, another most important thing,

488

:

which I mean, not most of it is one

of the, his lessons learned, right?

489

:

Uh, because.

490

:

I have al always understood,

I mean, always thought, right?

491

:

By viewing the, the, these kinds of, you

know, human, uh, uh, uh, you can say, man,

492

:

uh, created artifacts as systems, right?

493

:

Systems gives you a, an ontology of,

you know, components, connections.

494

:

They are evolving over time.

495

:

They're spatially, you

know, kind of thing.

496

:

All of that semantics, you know, gives

a, a, a, you can say a, a, a, a, a good

497

:

way to exploit that semantics, right?

498

:

The causality in a systems world, right?

499

:

The dynamics in a systems world, right?

500

:

Some of this is not present in

what is in this thing, right?

501

:

But in order to bring that

richness of AI reasoning, right?

502

:

Whether I'm saying, Hey, why

should I be having this kind of a

503

:

duck at this kind of a thickness?

504

:

Hey, is it going to be a pressure drop?

505

:

I need that systems thinking of breaking

it up into, you can say subsystems, right?

506

:

Their components, their

relationships, right.

507

:

And then being able to reason about them.

508

:

So, so you are right.

509

:

Absolutely.

510

:

Uh, that's, uh, one of

the primary motivation.

511

:

So when we are sucking in our model,

uh, we obviously create an, a more

512

:

neutral representation that is more

augmented with these kinds of, uh,

513

:

you can say that the semantic aspect,

I mean system oriented aspects.

514

:

Uh, as well as decomposition.

515

:

It's also a fundamental to how we scale.

516

:

Right.

517

:

Uh, in terms of, if you need, I'm just

going, going in a little bit again,

518

:

deeper into why, rationalizing, why

do I need systems that if I'm going to

519

:

basically think I'm gonna mirror the

world of the architects, the builder,

520

:

the, the, the guy who's doing plumbing

or the MEP engineer, they work in

521

:

their quote unquote subsystems work.

522

:

They have relationship, right?

523

:

So the notions of concurrency that

asynchronously each one working

524

:

present, so does the, is the, is the

representation designed like that

525

:

that allows you to, you know, work

in there, you know, independently

526

:

and synchronize and collaborate and

cooperate where there are connectivity.

527

:

And so that's another important

thing, which basically is like saying

528

:

by design, we are creating a system

that is geared for reasoning in, you

529

:

know, by multiple agents as surrogate.

530

:

They can do teamwork, they can,

you inference, they can find out

531

:

what's affecting what, right?

532

:

So, so those, you know, you can say again,

you, when you think about architecting an

533

:

intelligent, uh, system and you want to

see, hey, this world is a physical system,

534

:

I better represent that in that manner.

535

:

Then I can be actually.

536

:

Literally, you can say proactively

thinking future wise of Twinmaster.

537

:

It is by design.

538

:

By design has some properties that will

help it to grow and become more and more

539

:

effective.

540

:

Evan Troxel: I I have a question

about, about the kind of fundamentals,

541

:

'cause Michael said something and

then you've, you've all added to it,

542

:

uh, about thinking about Twinmaster

as a company from a new perspective.

543

:

Uh, because of you, you saw the technology

in the, in the future that was coming

544

:

and you wanted to build something

that could leverage that in a new way.

545

:

And then I think about architecture

firms and construction firms and

546

:

engineering firms that are very legacy.

547

:

This is the way we've always done

it, and I think about incentives.

548

:

And so this.

549

:

Randall, you can cut me off if

you think this is another podcast.

550

:

Okay.

551

:

'cause it might be.

552

:

Um, but, but the, the, you know,

incentives for architects and engineers is

553

:

to build a set of construction documents.

554

:

It's not to build a, an operations model,

which I think is more what you're talking

555

:

about, um, when it comes to digital twins.

556

:

Right.

557

:

That totally makes sense.

558

:

Right.

559

:

We're, we're talking about getting a

building permit based on abstracted

560

:

two dimensional drawings from

this three dimensional model.

561

:

And then architects are done,

you know, as soon as occupancy

562

:

happens and next project.

563

:

And then the owner is really the

one kind of, you know, facilities,

564

:

operations and management is taking over.

565

:

And maybe they're the ones who are,

obviously there's maybe a different

566

:

business model in the future that

architects could leverage, but for the

567

:

most part probably aren't going to.

568

:

And so therefore, you know, the

information that's stored in their models

569

:

may not be as conducive to the things

that you're talking about as ideal.

570

:

Right, because it's like, well,

the goal here is to get drawings

571

:

and then we're just going to,

we're going to build the thing.

572

:

We're gonna use that as a set of

instructions to build the thing.

573

:

And it's not maybe the same model

as as a BIM model during design.

574

:

Maybe it is, maybe it's augmented.

575

:

I don't know.

576

:

But I'm just curious, like from

your perspective, when it comes to

577

:

the way architects and engineers

do things, do you see enough of

578

:

what you did with Twinmaster, which

is rethinking the company or the

579

:

process of how we do this based on

the future technology that we see?

580

:

Dr. Prasanta Bose:

581

:

and it's,

582

:

Michael Jansen: Let's both

answer that, starting with your,

583

:

that's an excellent question.

584

:

It's a com, it's a

585

:

Dr. Prasanta Bose:

586

:

No, this, this is an excellent

question because this is, I have kept

587

:

on drilling into it with, with every

day, one hour, one hours, two hours.

588

:

I kept on and it, it, and it is, I, I

mean, I, I didn't know Frank Gary to be

589

:

honest with you, but when I read his, his

works, the way he worked, it gave me a

590

:

strong validation of the thinking that is.

591

:

If you design, you would, you know, and

this is also came for, uh, again, you

592

:

know, uh, out of Lockheed Martin, right?

593

:

When you're creating this multi-billion

dollar very, you know, cost intensive,

594

:

mission critical safety critical

system, you better design for that.

595

:

That means it is projecting into

the future, just as I were talking

596

:

about mal predictive sense, but in

an operational and in histories,

597

:

in a, in the future sense, right?

598

:

You need to basically, is this feasible?

599

:

Is this guy going to be, you know,

be even be buildable by cost?

600

:

Is the material there, is it

basically, uh, you know, uh, taken

601

:

care of, uh, with respect to,

uh, you know, these regulations?

602

:

The reason you wanna do that, and

again, this is lessons learned.

603

:

I heard the hardware right.

604

:

I was asked this backtracking, you

do not want to backtrack and fix very

605

:

costly mistakes in your design that is

going to bite you during construction

606

:

phase, during operations phase.

607

:

Right?

608

:

So this forward thinking, right?

609

:

I mean, you know, you're

thinking ahead of this thing.

610

:

You can fold in.

611

:

You are just like front, you know,

people use the word called front

612

:

loading or whatever you want to call it.

613

:

But being able to take that into

consideration, I mean, obviously you

614

:

are going to go in a lazy manner, right?

615

:

And you hear this economic story that

is they, they call it, as you know,

616

:

that is you don't over-engineer.

617

:

You know what I mean?

618

:

So it's like this options

market kind of thing.

619

:

You know which options you want to

exercise that is good enough, right?

620

:

But yet gives you that 70 or 80% this

thing so that you are really protected.

621

:

Okay.

622

:

And that's what has

been the central pieces.

623

:

I mean, of my, you can say training.

624

:

I try to train others, you know,

kind of in that thinking of, you

625

:

know, when you are designing, when

you are creating the twin all at the

626

:

early stages, think like that, right?

627

:

And, and so when you are simul,

you know, when you're saying,

628

:

Hey, what is the energy demand?

629

:

You're saying, Hey, what will be the,

the occupancy of this building, right?

630

:

What kind of a environment

is this guy located?

631

:

You know what I mean?

632

:

What are the buildings that

may possibly grow, right?

633

:

You are modeling your, basically, you

know, you're, you're doing an active

634

:

model construction, and based on that

you're saying, Hey, wait a minute,

635

:

based on, you know, what the, the,

the current site, the, the project is

636

:

right, and the environment around it.

637

:

These are, you know, you

can say impacts, right?

638

:

And this is what you're going to

make dishes based on that effects.

639

:

Michael Jansen: uh, I'm an

architect by training myself.

640

:

I, I spent 11 years in the profession,

um, five years with, uh, a company

641

:

called Portman and Associates that got

sold recently after many, many years.

642

:

But back in the day it was big into

hotels and mixed use developments.

643

:

All, they're driven by Mr.

644

:

Portman himself.

645

:

And then I went to, uh, on to the

interior side and joined Hirsch, Bedner

646

:

for several years and began working

on luxury hotels all over the world.

647

:

So I got a lot of, um.

648

:

Of hands-on project and construction

experience and global projects that

649

:

were upscaled typically city center

things and learned all the problems

650

:

architects faced from concept design to

delivery to permitting, to translation

651

:

on site, to having to translate things

across borders and having to work

652

:

with design institutions that were

responsible for making construction,

653

:

drawing, I mean all those issues.

654

:

I think when we created Twinmaster,

just a few points I wanna make.

655

:

One is that primarily the goal

was to try to make AI easy to

656

:

adopt for architecture firms.

657

:

Um, simple natural language prompts.

658

:

No need for scripting, no need for

additional plugins, and allow them to

659

:

keep using the tools they were using.

660

:

That was fundamental to us too.

661

:

To us.

662

:

'cause we saw a lot of other things

out there that were saying, Hey, try

663

:

my new platform and you have to learn

this and you have to learn that.

664

:

And there was a lot of pushback

against that from incumbent firms

665

:

saying, we just don't want to switch.

666

:

'cause we're, as you

said, our job is to get.

667

:

High quality projects out the door

that are well documented and done

668

:

on time to a certain standard.

669

:

I think the other thing we were

interested in, in terms of the process

670

:

itself is in trying to improve that,

you know, the, the trade off process.

671

:

I mean, architects are constantly

evaluating this versus that.

672

:

And so this concept of multi objective

reasoning was designed to make the design

673

:

process fast, accurate, and comprehensive.

674

:

So you could actually consider a

multitude of things simultaneously,

675

:

quickly and accurately.

676

:

Um, which as we know, I, I was a, I

started on hand drawings, I predate cad

677

:

and I went from, I was, because I was a

young guy at the time, they all, they took

678

:

all the young guys at Portman and made

us learn CAD and teach the older guys.

679

:

That was just what happened

to happen back in the day.

680

:

So I started at a micro station and

eventually moved to, to AutoCAD and so

681

:

then I founded a company called Sier

that provided CAD and BIM services to

682

:

what ended up being the 30th, the top 50.

683

:

AEC firms in the world.

684

:

And we had to have all these

different tools in the office.

685

:

'cause these companies were

using all these different tools.

686

:

So we had a team on MicroStation,

we had a team on, on AutoCAD.

687

:

We eventually switched to Revit

and by:

688

:

services, all that kind of thing.

689

:

So it became very familiar with the, uh,

the challenges of the process, and we just

690

:

really wanted to make it more efficient.

691

:

Um, and that kind of, we had this

idea that eventually it became ai.

692

:

And when I met presenter, I said,

finally, we have to, we have to

693

:

figure all this out and, and, and

see how we can apply what, you know,

694

:

to the building sector in a way that

architects and engineers can understand.

695

:

So there is a, a practical application

of this in the design process itself that

696

:

makes cuts planning time significantly,

that reduces WeWork significantly.

697

:

That makes, uh, that, that

reduces the need for micro

698

:

simulations dramatically and can a

699

:

Evan Troxel: you're pulling those

later, what, what maybe we would, we

700

:

would've considered later stage things

earlier into the design process so

701

:

that they have impact on the outcome,

702

:

Michael Jansen: A hundred

703

:

Evan Troxel: and then it can go farther

704

:

Michael Jansen: Absolutely.

705

:

Which leads to my third

point, an excellent segue.

706

:

I do believe the architects need to

think about other business lines.

707

:

And one of the things I think that

they can eventually, uh, consider is

708

:

being in the digital twin business

because they actually give birth

709

:

to the original digital twin model.

710

:

And I think there's a, there's a

business to be created out there that

711

:

would be substantial for architects

to offer digital twin as a service.

712

:

'cause they actually developed the

twin in the first place to the owner

713

:

that can be sold in perpetuity as,

uh, a residual of some kind that they

714

:

could manage, that they continuously

update, all that type of thing so

715

:

that they can move beyond the world of

fixed fees into a world of residuals.

716

:

I think we need to see architects

going that way into, you know,

717

:

rethinking how they make money.

718

:

We've all seen how.

719

:

Our fees have been cut because little

contractors take it more because of Ben.

720

:

The owner keeps pressing this down and

it's squeezing us to the point where

721

:

it's getting hard to do our work.

722

:

Architects work crazy hours 'cause

they're under constant pressure

723

:

to get jobs out within a certain

timeframe to a certain quality.

724

:

That's not gonna change.

725

:

But I think what can happen is they can

begin to in to spin off new business

726

:

lines as a result of the intellectual

property that they're creating.

727

:

So I'm hoping that we can put AEC firms,

specifically architecture engineering

728

:

firms into the digital twin business

someday, and that's one of our goals.

729

:

Dr. Prasanta Bose:

730

:

I, I, I think, uh, to add to that,

Michael, it's a, it's a, it's an

731

:

actually an opportunity for growth

of architects and engineers.

732

:

Okay?

733

:

I'll tell you why.

734

:

Uh, again, if you think about the

complexity of the world, okay,

735

:

in which this, this building is

gonna be placed, they are, are

736

:

constantly, you know, confronted

with articulating their questions.

737

:

Obviously, in the world, in the parliament

of Prompt as a framework, they have

738

:

to articulate, you know what I mean?

739

:

Hey, what's the problem?

740

:

You know, why does it impact this?

741

:

They're constantly rationalizing

the distance from this

742

:

multiple dimensions, right?

743

:

So in doing so, right, uh, it, it is

not actually, uh, you, you can say, uh,

744

:

giving away the knowledge to an agent,

but taking that and act, you know,

745

:

getting these agents help them out.

746

:

They have just now make

it a first class concern.

747

:

You see what I'm saying?

748

:

That is now present in the kind of, you

can say the reasoning that went behind

749

:

it, the stateful, you can, whatever, you

know, uh, uh, scene representations, the

750

:

causal graphs being created, that is now

they can use for the subsequent answers.

751

:

They're becoming more themselves educated.

752

:

Right.

753

:

And the products they're creating

as a result of that is going to

754

:

make them more, more powerful.

755

:

I mean, how would I say?

756

:

They, they will unlock

their creativity much

757

:

Randall Stevens: It is the, ultimate

manifestation of the master architect.

758

:

The architect wants to be co, I want,

I wanna control the entire context,

759

:

and that's what you're saying.

760

:

Give me the entire context window,

761

:

and.

762

:

Yeah.

763

:

Dr. Prasanta Bose:

764

:

Yeah.

765

:

I mean, I always think, and.

766

:

this is the thing I I've been, I've

been, you know, I'm becoming more,

767

:

more of a preacher that actually,

you know, it, uh, it, it, it just

768

:

unlocked you, you know what I mean?

769

:

Now you have the power without

the emotion, without the thing to

770

:

ask the relevant question and give

shape to your thinking of the thing

771

:

Randall Stevens: No, I think

it makes a lot of sense.

772

:

I think, and I think it's what

the skill of an architect is.

773

:

They, they have a, there are

many, there are many factors

774

:

that go into making decisions about what

775

:

Dr. Prasanta Bose:

776

:

Yes, yes.

777

:

I mean, they, they're confronted with it.

778

:

They, they don't have a choice.

779

:

You see what I'm saying?

780

:

So now you have something is coming

about to in front of them, like

781

:

twin Maie that is helping them

to, the friction is going away.

782

:

You know what I mean?

783

:

The cognitive gap is gone right now.

784

:

They can express and say, ah, I see,

you know, if I design it in this way,

785

:

if I get this kind of a curvature,

I'm gonna get this kind of Right.

786

:

It's, it

787

:

just changes the way you work.

788

:

It just transforms

789

:

Evan Troxel: the design process is

a relationship with the client until

790

:

it becomes a transaction, and then

we're gonna do that all over again.

791

:

But the thing that you're, you're

talking about, and I've talked

792

:

about this previously on the Troxel

podcast, is like this, a potential

793

:

to continue the relationship, right?

794

:

And

795

:

Dr. Prasanta Bose:

796

:

That's what Michael said.

797

:

Exactly.

798

:

Evan Troxel: insight

and help those clients

799

:

Dr. Prasanta Bose:

800

:

Yeah.

801

:

It's like your, it, it is like, yeah.

802

:

It's basically what you did was you

embodied your intelligence into the

803

:

twin and it's carrying its journey.

804

:

Okay.

805

:

Into the construction.

806

:

It is, you know, becomes,

now I don't have to be there.

807

:

I already have given that,

you know what I mean?

808

:

When it is in the design, he,

this guy's gonna talk for it

809

:

is enough on behalf of me.

810

:

It's like that, you know, whether

it's like the, in the construction,

811

:

in the operations or whatever,

maybe the, you know, the, the

812

:

lifecycle

813

:

Randall Stevens: I think it makes sense.

814

:

I think Michael, you're, you're right.

815

:

It's like the opportunity to create a new

revenue generating business model for the

816

:

architects that are out there to say, you

know, we, we shouldn't be as transactional

817

:

as we have been in the past, and can

we have a deeper, longer relationship?

818

:

And, you know, if this becomes part

of, part of embedded in that twin is.

819

:

Is you right.

820

:

As part of carrying that

information forward.

821

:

Yeah,

822

:

Dr. Prasanta Bose:

823

:

The soul, the soul of the architect.

824

:

Randall Stevens: so I've got a,

uh, somewhat technical question

825

:

and then, and then we'll be a

little bit pragmatic about, uh, the

826

:

Twinmaster implementation of this.

827

:

But when you, um, Presanta when you were

talking about the, the, the data coming,

828

:

say geometry from one of these systems,

uh, into this neutral ground in order

829

:

to, to use the AI agent, uh, or, or

do do this analysis on think something

830

:

as simple as like a wall is just two

two planes, whereas in reality that

831

:

is a composite complex system, right.

832

:

With a lot of things.

833

:

So how do you, um, you know, if

you are bringing geometry from

834

:

one of these systems over, are

you basically tagging those with.

835

:

It better, deeper information

that you don't have to model it.

836

:

You don't have to re, in reality, you

don't have to physically model everything.

837

:

You just need to reference

838

:

what this thing might look like.

839

:

What, what is, how do you do that?

840

:

What

841

:

Dr. Prasanta Bose:

842

:

So it boils

843

:

down to, uh, again, in the world of,

so when you think about common sense

844

:

reasoning, most of the time common sense

reasoning about is some, some kind of

845

:

a semantic framework, an ent, right?

846

:

What's a wall?

847

:

You know, it has occupied space,

it is had materials, it has

848

:

some structure in it, right?

849

:

All of these things comes

in right into the picture.

850

:

So when I'm bringing into this

thing, uh, the, into my thing,

851

:

first and foremost again.

852

:

Going back to my, uh, this, this idea

of, of a system that there's a, you know,

853

:

the whole system, like let's say the

structural system or the architectural

854

:

subsystem, right, which has component of

parts in a physical sense parts, right?

855

:

Uh, and their relationship or each

part is now having a more deeper.

856

:

Uh, information and it grows, keep in

mind, right, because what it starts as

857

:

journey at the, let's say in the Revit,

uh, environment or in acad, when I

858

:

bring it in, uh, I give more meaning.

859

:

But as it grows into, let's say

I'm doing quote unquote the carbon

860

:

footprint, I am now putting causality

of saying that wall situated in this

861

:

particular floor with this kind of an

environment is causing X, Y, z heat loss.

862

:

I'm just getting some sense of this.

863

:

Right?

864

:

So that's the meaning.

865

:

It's taking more and more meaning

down the line so that now you can

866

:

say, Hey, wait a minute, I'm using

this material and this causing this.

867

:

Right?

868

:

Okay.

869

:

Randall Stevens: are, you mixing?

870

:

Like, could that be, uh, you

know, a geometric representation,

871

:

but then also a, a document that

872

:

describes the properties

of that material that

873

:

Dr. Prasanta Bose:

874

:

It, it, it, uh,

875

:

Randall Stevens: connected to each.

876

:

Dr. Prasanta Bose:

877

:

It, is de definitely the, uh, right.

878

:

So, uh, in the, in the, I won't

say the good old days of the, I

879

:

mean, so if you, if you look at

ai, AI has grown from symbolic ai.

880

:

This w the AI as a, you know,

if you look at ai, AI has a

881

:

multi-dimensional aspects of it.

882

:

Ai, there's this big, uh, you can

say group of people in the circa

883

:

19, uh, I mean, you know, nineties

to this, uh, called symbolic, yeah.

884

:

Representation.

885

:

The word is, you know, how do you, how

do you represent things of the world?

886

:

How do we represent things, right?

887

:

So there's this, you know, it's

all about semantics, right?

888

:

How do we do grounded, how do you

do common sense reasoning with it?

889

:

Okay.

890

:

So there's that aspect of it.

891

:

And then, you know, goes into other kinds

of things about, Hey, how about, uh,

892

:

you know, AI as search, AI as reasoning,

this kind of an operational sense, right?

893

:

So the representational side

of things in ai, people have.

894

:

But you know, PhD after PhD

has gone in there, right?

895

:

Which is how do I structure, right?

896

:

Uh, and you see this in this, in,

in, in, even in neural nets, right?

897

:

When you look at a, a, a complex construct

like a wall, what's, what's the, right,

898

:

what, how do you organize, you know,

there is different pieces of information.

899

:

Is it a monolithic document or is

it the, these, if I put a neural,

900

:

uh, head on it, it's like saying,

okay, I got a token for this.

901

:

I have a token for that.

902

:

I have a token for that.

903

:

And they're related in this manner.

904

:

Right.

905

:

Okay.

906

:

I'm just giving you Right.

907

:

Obviously if you boil it down, obviously

it'll have a binary vector, but at the

908

:

symbolic level, so when I'm looking

at wall and his representation, I

909

:

am looking at a little bit what they

call as a structured representation

910

:

rather than a clean document.

911

:

I could have done that.

912

:

I could have keyed a more description.

913

:

I do that when I'm facing the user,

that means I have to create a report or

914

:

something that's easy to generate, but

for my efficiency of reasoning, to be able

915

:

to generalize, to look at the causality,

what contributes this thing to make it

916

:

more efficient, I need a little bit of a

more of a graph representation of that.

917

:

There are multiple benefits to that.

918

:

Right.

919

:

And I'm, again, not going into

the pot to do the training and

920

:

learning that goes in there.

921

:

Right?

922

:

Because what I've done is I have

passed the representation into

923

:

these little pieces of information

I have basically, you know, talked

924

:

about a semantic aspect of it.

925

:

Okay.

926

:

And then there's a very important reason

why you can ask why do you wanna do that?

927

:

Because it's, at the end

of the day, if you look at.

928

:

How we were, it's all about

how would say context.

929

:

You know, what, where is my attention?

930

:

What is the semantic thing,

which is kind of giving you the

931

:

where I want to pay attention.

932

:

Is it the wall's structure?

933

:

Is it the wall as an insulation material?

934

:

Or is the wall as a, you know, as

a concrete mass as the cost, right?

935

:

So these are different kind of semantic

thing and it has a spreading activation

936

:

the way, you know, the encoders decoders

work to bring in, generate information.

937

:

So that's the reason why we, we represent

a more structure so that they are

938

:

easily amenable to grounded reasoning

939

:

in an AI sense.

940

:

Right?

941

:

And then furthermore, they're

easily amenable for tuning my AI

942

:

models.

943

:

Randall Stevens: I was gonna say do, so

are y'all training your own models with,

944

:

are you training your own models?

945

:

Dr. Prasanta Bose:

946

:

I, I, I, oh, so you could have, this is

another thing that we initially, very,

947

:

very early on decided be, uh, I, not yet,

I would say, um, you, I would use the word

948

:

more in a tuning sense and more in the,

uh, how would I say it is, like, uh, how

949

:

do I create more, you know, just like.

950

:

I'm now focusing more on what is

called as the, like, the what,

951

:

what philanthropic is doing.

952

:

Give me the context, make sure I got

the context right so that I can make

953

:

correct reasoning, more effective

reasoning, that kind of thing, right?

954

:

So that part of the contextualization,

I, you know, foc we are have put

955

:

a lot of effort in that for, for

multiple reasons because I want

956

:

to be focused on this goal, right?

957

:

What's the context in which I'm coopering?

958

:

Uh, so we are not going back to your

question, not creating our models, but

959

:

focusing more on the existing gel purpose

ones and creating contextualization

960

:

as a way to make it more tuned and

then compiling that contextualization

961

:

into the thing, into further

962

:

Randall Stevens: So that's a good

segue to what I was gonna ask second,

963

:

which is if you are then fine tuning.

964

:

What did you all end up focusing on first

as a target problem for the Twinmaster?

965

:

Could help to solve 'cause

you can't do 'em all at once.

966

:

So was there a specific

area that you all chose

967

:

and then, and why?

968

:

Dr. Prasanta Bose:

969

:

Uh, I, I mean, I, I think that's something

is, is, is in the works, I would say,

970

:

uh, in, in the sense we are, uh, because

we had to beg a conscious decision

971

:

initially that, uh, let's complete

the reasoning framework, but before

972

:

putting more effort on optimizing the

tuning process, you know what I mean?

973

:

Like, hey, which ones I need to,

what, what is the most value add

974

:

from an efficiency standpoint Okay.

975

:

Kind of thing.

976

:

Uh, and so a couple of things that we

are now have, have initially focused

977

:

is, is like more to do with creating

the, the, the graph structures.

978

:

That can feed into the tuning.

979

:

And so when I say graph structures is

basically saying like, do I have the

980

:

right, uh, so if you look at token, right?

981

:

Text is one single word as a token, right?

982

:

But I can create to tokens,

which are kind of a graph.

983

:

This is related to this by this, right?

984

:

So that I can generalize, okay,

for my causality that this causes

985

:

this, that causes that, right?

986

:

And therefore I'm going to

basically get a big cost hit, right?

987

:

So where the focus has been with,

for, for the tuning standpoint is

988

:

preparing the data that I can then

feed into this, you know, today's,

989

:

you know, open source things

that we have been looking into.

990

:

Randall Stevens: Right.

991

:

Michael Jansen: I can

add a little bit to that

992

:

too.

993

:

I think to, to your question, you, you,

you can't do everything at at once.

994

:

That's just, it's, um, from a, from a

production perspective, it's inefficient

995

:

and all of sudden you can't assume too

much without having real experience

996

:

on real projects and real firms.

997

:

So we did make some

assumptions in the beginning.

998

:

We focused on certain, uh, performance

metrics before others, and now we're

999

:

focusing as, as said, on tuning and

optimizing and trade-offs among those.

:

00:51:25,142 --> 00:51:26,577

And where we're at is.

:

00:51:27,277 --> 00:51:31,177

We kind of have what I call a mature

MVP if that, if that is an oxymoron.

:

00:51:31,267 --> 00:51:35,137

Um, and we're at that stage where

we're putting into the hands now of,

:

00:51:35,197 --> 00:51:39,547

of, of all kinds of firms, big and

small, to give us their feedback so

:

00:51:39,547 --> 00:51:41,407

we can further tune our priorities.

:

00:51:41,407 --> 00:51:47,407

But we started off with energy,

carbon compliance, constructability,

:

00:51:47,857 --> 00:51:54,907

cost clash detection, daylight risk

and radiance, and now a new one

:

00:51:54,907 --> 00:51:55,147

that just

:

00:51:55,147 --> 00:51:56,377

got added thanks to chaos.

:

00:51:57,547 --> 00:51:59,677

Well, there's above 15 more

that we could be working with.

:

00:51:59,737 --> 00:52:00,057

Randall Stevens: lot though.

:

00:52:00,127 --> 00:52:00,417

Yeah.

:

00:52:00,692 --> 00:52:02,887

Michael Jansen: you, you could, yeah,

you couldn't do 'em all, but we couldn't

:

00:52:02,887 --> 00:52:06,997

just do one or two either because that

wasn't, uh, an adequate representation

:

00:52:06,997 --> 00:52:08,317

of the complexity of the problem.

:

00:52:08,377 --> 00:52:11,887

So working with chaos recently,

they've now added comfort into the

:

00:52:11,887 --> 00:52:14,587

mix, because that's a big concern for

how they're approaching the world.

:

00:52:15,097 --> 00:52:18,427

So we will, you know, we're going

through this process of evaluating

:

00:52:18,427 --> 00:52:21,247

what their priorities are,

because what I've found is that.

:

00:52:22,717 --> 00:52:25,717

Despite the way that, um,

architects design some of these

:

00:52:25,747 --> 00:52:28,957

fun, these considerations are kind

of fundamental to the practice.

:

00:52:28,987 --> 00:52:33,307

You know, so whether it's a fancy

Frank Gary building or it's a, uh, more

:

00:52:33,307 --> 00:52:38,497

mundane, uh, big box retail or hospital

building, there, there are these basic

:

00:52:38,497 --> 00:52:40,327

considerations that all buildings have.

:

00:52:40,567 --> 00:52:45,127

And so we're still going through a process

of, of studying and, and optimizing.

:

00:52:45,127 --> 00:52:46,627

And that's gonna continue

for a little bit.

:

00:52:46,807 --> 00:52:50,377

You know, I think we, we have to continue

to work with firms the way we've begun to

:

00:52:50,917 --> 00:52:52,807

and, and answer some of these questions.

:

00:52:52,807 --> 00:52:57,217

So I see we've got another, you know,

we're, we've been around for three years.

:

00:52:57,307 --> 00:53:01,147

Um, we're, we're at a

strong MVP at this point.

:

00:53:01,147 --> 00:53:04,417

I think it's probably another year

plus before we've solved some of

:

00:53:04,417 --> 00:53:08,497

these issues the way we want to,

we're calling, our formal launch is

:

00:53:08,497 --> 00:53:13,177

actually gonna be, we we're taking

on customers, but we're not really

:

00:53:13,177 --> 00:53:14,917

gonna be scaling up until next year.

:

00:53:15,097 --> 00:53:19,627

We're still going through this process

of, of making clients happy, getting

:

00:53:19,627 --> 00:53:21,307

their feedback, and incorporating it.

:

00:53:21,667 --> 00:53:26,467

And getting our, our, uh, software

partners, uh, content with what

:

00:53:26,467 --> 00:53:29,137

we're producing for them, getting

introduced into their partner,

:

00:53:29,197 --> 00:53:30,637

uh, and customer networks.

:

00:53:31,117 --> 00:53:33,547

And so there's still some work

to be done there before we really

:

00:53:33,547 --> 00:53:37,237

can put the pedal to the floor

and, and, and try to scale this.

:

00:53:37,477 --> 00:53:39,847

There's still some more lessons,

I think, to be learned for us.

:

00:53:39,847 --> 00:53:43,807

So, um, but the good news is that you

can get your hands outta the software

:

00:53:43,807 --> 00:53:46,267

today, use it and derive value.

:

00:53:46,327 --> 00:53:50,677

And so, um, that enables us

to, to, to form commercial

:

00:53:50,677 --> 00:53:52,057

arrangements with these companies.

:

00:53:52,057 --> 00:53:56,437

And we're now, um, in the process

with, with several of them, of

:

00:53:56,497 --> 00:53:59,467

engaging with their, their, their

customers for the first time.

:

00:53:59,467 --> 00:54:01,807

And it's, it's a learning

process right now, to be honest.

:

00:54:02,272 --> 00:54:05,012

Randall Stevens: So like, uh,

uh, you mentioned compliance,

:

00:54:06,807 --> 00:54:08,397

you know, and there's been a

couple of companies that have.

:

00:54:09,402 --> 00:54:11,502

Talked about trying to

put their arms around.

:

00:54:11,502 --> 00:54:16,032

All of that is your's approach

to let the customers bring their

:

00:54:16,032 --> 00:54:19,002

own data to, to bring into this.

:

00:54:19,002 --> 00:54:23,232

So you're not gonna, you're not gonna

be responsible for touching what are

:

00:54:23,232 --> 00:54:27,942

all the right code compliance, but

if somebody can connect, you know,

:

00:54:28,062 --> 00:54:33,792

point to or connect to wherever the

code compliance documents are, are

:

00:54:33,792 --> 00:54:35,472

you able to then bring that into the

:

00:54:35,472 --> 00:54:36,912

context model and, and

:

00:54:37,322 --> 00:54:37,323

Dr. Prasanta Bose:

:

00:54:37,323 --> 00:54:39,842

So yeah, I, I think

it's, it's a combination.

:

00:54:39,842 --> 00:54:40,382

I would say.

:

00:54:40,442 --> 00:54:46,652

Uh, one is, uh, uh, what, what, what

I call it as, uh, is like a kind of a

:

00:54:46,682 --> 00:54:50,972

gimme, a generalized con, uh, compliance

model, a generalized compliance model.

:

00:54:51,002 --> 00:54:54,932

That means, that means I have

structured it compo, you know, uh,

:

00:54:54,962 --> 00:54:56,462

it's like top down, bottom up, right?

:

00:54:56,702 --> 00:55:02,102

Top down means I can infer based on

these, uh, uh, on your building that

:

00:55:02,102 --> 00:55:04,292

these are the compliance components, okay?

:

00:55:04,292 --> 00:55:04,862

Number one.

:

00:55:05,162 --> 00:55:11,582

Number two is that if I then instantiated

or grounded based on the building, okay?

:

00:55:11,732 --> 00:55:12,722

On those components.

:

00:55:13,067 --> 00:55:13,397

Right.

:

00:55:13,397 --> 00:55:16,937

As I said, is a generalist, I'm

tuning that, those compliance things.

:

00:55:17,297 --> 00:55:18,737

But I wanna do two things.

:

00:55:18,797 --> 00:55:23,267

I want to get validation from the

user, number one, or validation based

:

00:55:23,267 --> 00:55:25,457

on his data about the compliance.

:

00:55:25,517 --> 00:55:25,847

Okay.

:

00:55:26,177 --> 00:55:28,277

So it's like top down, bottom up, right?

:

00:55:28,277 --> 00:55:29,237

You are grounding it.

:

00:55:29,242 --> 00:55:32,897

You, you are bringing in a, a

kind of a, a common framework, a

:

00:55:32,897 --> 00:55:37,247

schema as one would say that way

I'm not, uh, you know what I mean?

:

00:55:37,247 --> 00:55:39,497

I can always work with that schema.

:

00:55:39,707 --> 00:55:44,027

I can tailor it and customize

for this, uh, arch, uh, location

:

00:55:44,027 --> 00:55:47,927

and this, uh, architect and his

building versus that building.

:

00:55:47,957 --> 00:55:48,317

Okay.

:

00:55:48,617 --> 00:55:52,457

And that's been the kind of the general

framework that I've worked with.

:

00:55:52,457 --> 00:55:54,767

Uh, you know, this is again,

coming from lucky days.

:

00:55:54,797 --> 00:55:55,187

Okay?

:

00:55:55,547 --> 00:55:56,237

Lucky days.

:

00:55:56,237 --> 00:55:58,007

I mean, I'll give you an example.

:

00:55:58,307 --> 00:56:03,587

The people used to hide their, uh, you

can say all these engineering tools

:

00:56:03,587 --> 00:56:05,297

in these Excel spreadsheets, okay?

:

00:56:05,732 --> 00:56:06,992

I had no clue.

:

00:56:07,017 --> 00:56:11,102

I, I, I used to struggle to get

it out of the, you know, computer

:

00:56:11,282 --> 00:56:14,912

because I wanted to create a twin that

cuts across, you know, no crevices

:

00:56:14,912 --> 00:56:19,982

between, you know, uh, a mechanical

guy, a, a, a solar power guy, right?

:

00:56:20,372 --> 00:56:22,532

Uh, and those was no common framework.

:

00:56:22,802 --> 00:56:23,162

Okay?

:

00:56:23,372 --> 00:56:27,752

And that's why we are saying, Hey, if

I create a common stat, you know, uh, a

:

00:56:27,752 --> 00:56:33,632

generalizable framework or schema that

I'm generating, literally you can say

:

00:56:33,632 --> 00:56:36,542

dynamically on demand and design time.

:

00:56:36,542 --> 00:56:39,302

So there's this trade off

that you'll see in ai.

:

00:56:39,602 --> 00:56:41,822

And this is another thing that I

wanted to kind of share with you.

:

00:56:42,752 --> 00:56:45,032

This is, this is a, a

strong thinking that I have.

:

00:56:45,242 --> 00:56:47,882

There's a design time

AI versus runtime ai.

:

00:56:48,707 --> 00:56:49,127

Right.

:

00:56:49,307 --> 00:56:52,757

It is like, how many jewels are

you doing every time you hit and

:

00:56:52,997 --> 00:56:55,517

do something versus design time?

:

00:56:55,517 --> 00:57:00,767

You create an optimal kinda thing

and you continue to, you know, tune

:

00:57:00,767 --> 00:57:03,377

it, adopt it, configure it like that.

:

00:57:03,407 --> 00:57:03,767

Okay?

:

00:57:04,037 --> 00:57:07,637

That gives you delta energy cost, right?

:

00:57:07,817 --> 00:57:09,137

Maximum reuse.

:

00:57:09,227 --> 00:57:09,647

Right?

:

00:57:09,977 --> 00:57:14,867

And that's the thinking, which is that,

hey, I have, you know, I can sit on it, I

:

00:57:14,867 --> 00:57:19,187

can experiment, I can see how this thing,

and from a systems angle, this looks good.

:

00:57:19,247 --> 00:57:19,667

You know what I mean?

:

00:57:19,667 --> 00:57:20,297

It's solid.

:

00:57:20,447 --> 00:57:24,617

It has the, you know, flexibility,

it has the right components,

:

00:57:24,617 --> 00:57:26,177

the right modularity, right?

:

00:57:26,177 --> 00:57:27,467

These are engineering principles.

:

00:57:27,467 --> 00:57:29,927

I'm assessing that guy, right?

:

00:57:30,107 --> 00:57:32,087

And then I give it to the agent.

:

00:57:32,267 --> 00:57:36,107

And as the agent is, you know,

exercising it, it's tuning it.

:

00:57:36,797 --> 00:57:40,727

I mean, obviously for now, from a

reasoning standpoint, but further

:

00:57:40,727 --> 00:57:43,457

tuning it into the net, into

the, into the, into the, into

:

00:57:43,457 --> 00:57:45,257

the bid vectors, into the binary.

:

00:57:46,287 --> 00:57:50,157

Randall Stevens: So what's, uh,

what, what's the next thing that

:

00:57:50,157 --> 00:57:51,177

you think you're gonna tackle?

:

00:57:51,177 --> 00:57:51,447

What the

:

00:57:51,497 --> 00:57:51,498

Dr. Prasanta Bose:

:

00:57:51,498 --> 00:57:52,697

Oh, what's the next,

:

00:57:52,697 --> 00:57:53,087

so should

:

00:57:53,157 --> 00:57:53,367

Randall Stevens: You've

:

00:57:53,367 --> 00:57:54,507

got the MVP, so

:

00:57:54,677 --> 00:57:54,678

Dr. Prasanta Bose:

:

00:57:54,678 --> 00:57:55,637

let Michael say.

:

00:57:55,692 --> 00:57:57,967

Michael Jansen: We, we, we

both answered that differently.

:

00:57:57,967 --> 00:57:59,707

I think that Presanta could answer.

:

00:57:59,767 --> 00:58:02,977

Um, maybe you should first about

technologically and feature wise

:

00:58:02,977 --> 00:58:04,477

what you, what you wanna see happen.

:

00:58:04,477 --> 00:58:06,397

And I'll talk a little bit about

how I wanna see this scale.

:

00:58:06,397 --> 00:58:07,177

So, go ahead, Prisa.

:

00:58:07,197 --> 00:58:07,257

It.

:

00:58:07,782 --> 00:58:07,783

Dr. Prasanta Bose:

:

00:58:07,783 --> 00:58:12,762

I think a couple of things, uh, what's

next coming in is, you know, take this,

:

00:58:12,762 --> 00:58:20,082

uh, the, the, the digital twin and

Archie, uh, to do more of the spatial

:

00:58:20,082 --> 00:58:22,152

reasoning, the world of spatial reasoning.

:

00:58:22,272 --> 00:58:22,602

Okay?

:

00:58:22,782 --> 00:58:23,802

Because it's a physical world.

:

00:58:24,402 --> 00:58:28,092

So whether it is for, you know,

am I designing for safety?

:

00:58:28,362 --> 00:58:32,742

Is this guy the heat pump that

I done, right, because of the

:

00:58:32,742 --> 00:58:36,912

way I have, you know, set it

up, is there any pressure drop?

:

00:58:37,092 --> 00:58:42,252

So this is basically reasoning about

flows, reasoning about the spatial

:

00:58:42,252 --> 00:58:43,722

environment in which one, you know.

:

00:58:45,032 --> 00:58:47,132

Finds itself in a built space.

:

00:58:47,192 --> 00:58:47,522

Right.

:

00:58:48,032 --> 00:58:51,392

You know, so that will become

more and more one central theme.

:

00:58:51,812 --> 00:58:55,652

I mean, it'll, uh, it's, it's

where, uh, Archie will take

:

00:58:55,652 --> 00:58:57,632

more, you can say more meat.

:

00:58:57,782 --> 00:59:03,017

I would say the, the other thing that I've

been, you know, uh, thinking very hard.

:

00:59:03,877 --> 00:59:08,587

Or, or, you know, today when we

light up, that is, we always, the

:

00:59:08,587 --> 00:59:13,237

interaction is with your, the building,

but the building is now showing

:

00:59:13,237 --> 00:59:15,067

it in 3D You see what I'm saying?

:

00:59:15,277 --> 00:59:16,087

Whatever.

:

00:59:16,297 --> 00:59:19,837

You can say the, it's

assets, it's carbon, right?

:

00:59:19,837 --> 00:59:24,247

It's like the heat map in the, so

that the data, the, the inferences

:

00:59:24,247 --> 00:59:26,347

are anchored in the physical space.

:

00:59:26,347 --> 00:59:27,307

3D space, right?

:

00:59:27,607 --> 00:59:33,877

So, so today, the way we do that, uh,

you know, grounding is all great, but

:

00:59:33,877 --> 00:59:38,767

then, uh, there, there is also room

for giving you more finer fidelity.

:

00:59:38,797 --> 00:59:43,297

And this is the, you hear this call

gian splattering and all things because

:

00:59:43,297 --> 00:59:46,807

it is like this point clouds, which

gives you more higher resolution.

:

00:59:46,807 --> 00:59:48,367

It gives you a realistic feeling.

:

00:59:48,667 --> 00:59:50,527

So that's another direction it'll take.

:

00:59:50,887 --> 00:59:55,477

But then there are other ones that

are, uh, that I don't know how

:

00:59:55,477 --> 01:00:01,267

much to say, but the idea that

I'm after is that, uh, is there's

:

01:00:01,267 --> 01:00:02,652

a lot of built spaces, right?

:

01:00:02,682 --> 01:00:03,817

Kind of thing, right?

:

01:00:04,117 --> 01:00:09,877

and when I go into a building, uh, just

as I form my model of the building, I

:

01:00:09,877 --> 01:00:14,257

can immediately get ideas of how I can

change, Hey, hey, if I change this and

:

01:00:14,257 --> 01:00:16,597

this, it will get a different flow, right?

:

01:00:17,047 --> 01:00:20,917

So my question, or you can say a

challenge problem that architects

:

01:00:20,917 --> 01:00:24,427

face is, am I gonna do the whole thing

from scratch and redo this thing?

:

01:00:24,742 --> 01:00:31,012

No, what I want to do is Archie become

more powerful and you can say, I'll

:

01:00:31,012 --> 01:00:36,202

use the word reverse, engineering the

design right outta what he just saw.

:

01:00:36,742 --> 01:00:41,332

Archie got some pictures it reverse

engineered, get in them all for

:

01:00:41,332 --> 01:00:43,762

you, and it gave you the options.

:

01:00:43,882 --> 01:00:47,807

So that's the kind of the direction,

you know, few or three directions

:

01:00:47,997 --> 01:00:50,602

that, that, uh, Archie is gonna take.

:

01:00:50,602 --> 01:00:56,242

But definitely, uh, I think, uh, putting

Archie, I mean, center in front immediate

:

01:00:56,662 --> 01:01:01,552

future is, is really get this thing

into the, solve some hard problems

:

01:01:01,552 --> 01:01:03,682

this architects are facing, right?

:

01:01:03,742 --> 01:01:06,442

I mean, make, make them a,

a good assistant, a good

:

01:01:06,442 --> 01:01:07,132

companion.

:

01:01:07,282 --> 01:01:10,642

Randall Stevens: what you just, uh,

described is, uh, you know, I've

:

01:01:10,642 --> 01:01:15,562

been saying it a lot lately with, uh,

interactions with AI that, uh, it's

:

01:01:15,562 --> 01:01:17,872

definitely easier to edit than create.

:

01:01:18,502 --> 01:01:22,492

And one of the, the nice things about

ai, and I think what you were just

:

01:01:22,492 --> 01:01:25,252

describing was why start from scratch?

:

01:01:25,282 --> 01:01:27,052

You don't wanna start from the basics.

:

01:01:27,082 --> 01:01:29,032

Give me something and then

:

01:01:29,032 --> 01:01:31,642

I'll go to, I'll, I'll, I can go to town,

:

01:01:32,012 --> 01:01:32,013

Dr. Prasanta Bose:

:

01:01:32,013 --> 01:01:32,917

I'll, I'll message.

:

01:01:33,262 --> 01:01:35,752

Randall Stevens: shape it, and,

uh, I'm, I mean, I'm seeing it.

:

01:01:35,752 --> 01:01:40,372

I'm using, um, you know, I'm using

chat GPresanta lot, uh, just in

:

01:01:40,372 --> 01:01:42,952

trying to get the bus, the, the.

:

01:01:43,507 --> 01:01:47,287

The business model, the what are we

trying to do, you know, all the way from

:

01:01:47,287 --> 01:01:50,677

what does the product look like to, how

do you describe what the problem is?

:

01:01:51,157 --> 01:01:56,497

And it's, it's amazing to work

with these new tools because, you

:

01:01:56,497 --> 01:01:57,997

know, I'm, I'm one of these guys.

:

01:01:57,997 --> 01:02:02,137

I, I feel like I, when I'm around somebody

like you, Presanta, I, I realize how

:

01:02:02,167 --> 01:02:06,427

much I don't know, but I always consider

myself a reasonably intelligent guy.

:

01:02:06,427 --> 01:02:09,217

But it's like, you start interacting

with this and you're like, oh my

:

01:02:09,217 --> 01:02:11,437

gosh, it is smarter than me already.

:

01:02:11,887 --> 01:02:16,837

Uh, you know, in a, in a very generalized

way because it can take a large con,

:

01:02:16,867 --> 01:02:22,357

a large amount of context and begin to

put it into a form that is digestible

:

01:02:22,357 --> 01:02:24,997

by me that I can respond to and shape.

:

01:02:25,222 --> 01:02:25,223

Dr. Prasanta Bose:

:

01:02:25,223 --> 01:02:26,062

so, yeah.

:

01:02:26,062 --> 01:02:28,672

So I'm gonna just so Randall,

I'll, I'll, I'll make a

:

01:02:28,672 --> 01:02:29,272

correction there.

:

01:02:29,272 --> 01:02:33,682

Randall, it is, not that it's

smaller than you, what we grew up

:

01:02:33,712 --> 01:02:36,292

in, siloed access to information.

:

01:02:36,472 --> 01:02:36,862

Okay.

:

01:02:37,222 --> 01:02:39,982

You know, so I always

look at my growing, right?

:

01:02:40,012 --> 01:02:43,942

I used to use it microfiche and digging

into narrow this thing, scripts, right?

:

01:02:44,152 --> 01:02:48,352

Then came computers, then came

internet, then came damel, you

:

01:02:48,352 --> 01:02:50,002

know, semantic description, right?

:

01:02:50,152 --> 01:02:52,102

All on and on til ai, right?

:

01:02:52,252 --> 01:02:59,152

So as each one, you know, evolution

came access to information, right?

:

01:02:59,152 --> 01:03:00,202

Became easier.

:

01:03:00,442 --> 01:03:03,922

That means I can quickly

get it based on my intent.

:

01:03:04,042 --> 01:03:05,572

If you look at, that's

what Archie is doing.

:

01:03:06,082 --> 01:03:08,902

So what was not possible?

:

01:03:08,932 --> 01:03:11,272

What was kind of a siloed, right?

:

01:03:11,272 --> 01:03:12,952

And you had to really struggle.

:

01:03:13,402 --> 01:03:17,362

It gave you, if you asked the right

question, you are asking the question.

:

01:03:17,362 --> 01:03:18,287

You are always smart already.

:

01:03:18,622 --> 01:03:22,732

It is just that you did not

have the power to get to

:

01:03:22,852 --> 01:03:23,142

Randall Stevens: Yeah.

:

01:03:23,147 --> 01:03:23,497

Yeah.

:

01:03:23,587 --> 01:03:23,917

No, you're

:

01:03:23,932 --> 01:03:23,933

Dr. Prasanta Bose:

:

01:03:23,933 --> 01:03:24,592

now.

:

01:03:25,132 --> 01:03:25,942

Exactly.

:

01:03:25,942 --> 01:03:28,492

Now it operationalized information.

:

01:03:28,612 --> 01:03:30,352

It percolated, you know what I mean?

:

01:03:30,442 --> 01:03:33,172

It basically, they call as an,

you know, the small network.

:

01:03:33,352 --> 01:03:36,892

It just, you know, penetrated the

network, brought the information to

:

01:03:36,952 --> 01:03:39,952

you in a, in a cogent and relatable

:

01:03:40,117 --> 01:03:40,417

Randall Stevens: Yeah.

:

01:03:40,422 --> 01:03:41,647

And, and it's good.

:

01:03:41,647 --> 01:03:44,497

It, you know, the f the

first prompt is pretty good.

:

01:03:44,887 --> 01:03:48,757

I think, you know, it's gonna be people,

you know, giving them then the control

:

01:03:48,757 --> 01:03:52,177

to shape that and to do something with

it, personalize it, make it their own.

:

01:03:52,177 --> 01:03:53,767

And, uh, yeah.

:

01:03:53,767 --> 01:03:56,437

I think we're at the, uh, you

know, I, we're at the very

:

01:03:56,437 --> 01:03:58,597

beginning of a whole new wave

:

01:03:58,597 --> 01:04:00,367

of what this is gonna look like.

:

01:04:00,367 --> 01:04:00,697

Right.

:

01:04:00,697 --> 01:04:00,967

And

:

01:04:00,967 --> 01:04:02,767

Michael Jansen: if, uh,

AI were an alphabet, we're

:

01:04:02,767 --> 01:04:05,137

still at a, in AEC industry,

:

01:04:05,262 --> 01:04:05,263

Dr. Prasanta Bose:

:

01:04:05,263 --> 01:04:05,952

That's right.

:

01:04:06,012 --> 01:04:06,732

That's right.

:

01:04:07,027 --> 01:04:09,247

Michael Jansen: I just some practical

next steps for us for the rest of,

:

01:04:09,397 --> 01:04:10,567

throughout the rest of the year.

:

01:04:10,597 --> 01:04:15,067

Um, we have about 75 AEC firms

in our beta program right now.

:

01:04:15,547 --> 01:04:17,677

Mostly A, some E, some c.

:

01:04:18,442 --> 01:04:22,252

Varying sizes from tens of

thousands of employees to 200.

:

01:04:22,342 --> 01:04:25,462

You know, so they're all, we're

getting feedback from a lot of folks.

:

01:04:25,462 --> 01:04:30,172

So one of our goals is to continue

to learn from and with them to

:

01:04:30,172 --> 01:04:31,672

satisfy the needs that they have.

:

01:04:31,672 --> 01:04:34,462

We're getting a lot of questions

at this stage, which are, can

:

01:04:34,462 --> 01:04:35,572

it do this, can it do that?

:

01:04:35,572 --> 01:04:36,172

Questions?

:

01:04:36,712 --> 01:04:39,862

So that's an important part of

the, the growth process for us.

:

01:04:39,862 --> 01:04:43,552

What's lovely about working with

architects and engineers in particular

:

01:04:43,552 --> 01:04:45,232

is that they're in creative professions.

:

01:04:45,622 --> 01:04:47,572

So we get a lot of creative

suggestions from them.

:

01:04:47,662 --> 01:04:51,202

And, you know, I, I must say that's

been very helpful to us for sure.

:

01:04:51,892 --> 01:04:57,592

Um, as we satisfy the needs of these

firms, um, one of our goals is to

:

01:04:57,592 --> 01:05:02,512

build up a portfolio of, of, of case

tudies so that as we get into:

:

01:05:02,992 --> 01:05:06,622

we can, you know, very practically

show people that we did this on this

:

01:05:06,622 --> 01:05:08,062

project, we did that on this project.

:

01:05:08,062 --> 01:05:09,022

This is how you apply it.

:

01:05:09,022 --> 01:05:09,832

These are the results.

:

01:05:09,832 --> 01:05:10,702

This is what you would've done.

:

01:05:10,702 --> 01:05:12,352

This is what you do

now, that type of thing.

:

01:05:13,297 --> 01:05:17,737

Then, uh, simultaneously we wanted

to set up these new AEC firm partners

:

01:05:17,737 --> 01:05:21,487

for success, which means we're

actively working with the ones that

:

01:05:21,487 --> 01:05:26,077

I've already described to you, um, to

integrate with their existing tools.

:

01:05:26,467 --> 01:05:29,407

And they have structured

processes, all of them.

:

01:05:29,407 --> 01:05:33,457

It takes many months to get through

this process from identifying if there's

:

01:05:33,457 --> 01:05:36,757

interest, bringing the right people

around the table, beginning the technical

:

01:05:36,757 --> 01:05:40,297

integration, evaluating the technical

integration, testing it with their

:

01:05:40,297 --> 01:05:42,277

customer base, and then it goes to market.

:

01:05:42,277 --> 01:05:45,577

So we're in various stages of

that process with a number of

:

01:05:45,577 --> 01:05:47,017

companies, and that's gonna

:

01:05:47,442 --> 01:05:50,022

Randall Stevens: So do you end up,

do you end up wrapping, you know,

:

01:05:50,022 --> 01:05:51,552

just from a business standpoint?

:

01:05:51,702 --> 01:05:55,302

Uh, 'cause a, a lot of what, uh,

so I, I've run this company, uh,

:

01:05:55,482 --> 01:05:58,122

which is like a content management

platform that's being used.

:

01:05:58,122 --> 01:06:04,422

And, you know, what we found is that, um,

you know, change management inside of an

:

01:06:04,422 --> 01:06:06,282

organization is the, is the hard part.

:

01:06:06,312 --> 01:06:08,562

The people part is the hard

part, not the technology.

:

01:06:08,862 --> 01:06:13,212

So do you end up, um, kind of

wrapping, I'll just say professional

:

01:06:13,212 --> 01:06:15,132

services around what you're doing?

:

01:06:15,132 --> 01:06:18,792

Or is that a way that you're thinking

about, 'cause you're talking about

:

01:06:18,792 --> 01:06:22,842

changing the way that they're going

to design or use these tools to do

:

01:06:23,142 --> 01:06:26,082

their work every day, if or not.

:

01:06:26,408 --> 01:06:28,778

Michael Jansen: We're, we're, we're not

really Randall, we're not really going to,

:

01:06:29,018 --> 01:06:33,998

we have a, a small portion of our business

that will be professional services driven.

:

01:06:33,998 --> 01:06:38,018

And that's really for, 'cause there's

a small market out there for, um,

:

01:06:38,078 --> 01:06:42,368

standalone digital twin projects that,

um, are, are, are things at scale.

:

01:06:42,698 --> 01:06:46,958

And we will undertake those, um, on

a, on a, on a case by case basis.

:

01:06:47,468 --> 01:06:53,738

But primarily what we really want to do is

build out a replicable solution that is,

:

01:06:53,798 --> 01:06:58,628

uh, self-serve so that our, uh, partner

clients can, can run with it on their own.

:

01:06:58,898 --> 01:06:59,228

Right.

:

01:06:59,663 --> 01:07:02,723

And that's the process we go through

during the integration phase, during

:

01:07:02,723 --> 01:07:08,003

the testing phase, during the customer

evaluation phase, it's longer, but by

:

01:07:08,003 --> 01:07:10,613

the end of it, you've really proven

everything before you go to market.

:

01:07:10,673 --> 01:07:13,073

And that's kind of what you

have to do to be successful.

:

01:07:13,643 --> 01:07:16,733

Subsequent to that, you need to,

you know, there's gonna be some

:

01:07:16,733 --> 01:07:21,383

marketing involved because as you know,

there's, we can be in the Autodesk

:

01:07:21,383 --> 01:07:23,123

marketplace or the Trimble marketplace.

:

01:07:23,123 --> 01:07:24,353

That's a great place to be.

:

01:07:24,743 --> 01:07:28,943

But if you, again, look at the

hard numbers, 80% of the tools

:

01:07:28,943 --> 01:07:31,643

in those marketplaces don't

really get a lot of traction.

:

01:07:32,123 --> 01:07:33,773

So you really have to call attention.

:

01:07:33,773 --> 01:07:36,683

So the market knows that you're one of

them there and that you're, you know,

:

01:07:36,683 --> 01:07:39,248

that they know where to go find you

if they're a Rev user, if they're a

:

01:07:39,253 --> 01:07:41,183

SketchUp user, if they're an acad user.

:

01:07:41,663 --> 01:07:43,013

So there's all that to consider too.

:

01:07:43,043 --> 01:07:47,003

So there will be marketing behind

this that will be creating awareness

:

01:07:47,003 --> 01:07:49,103

of what RG brings to the table.

:

01:07:49,103 --> 01:07:53,093

That's gonna have to be built up too as

part of this year, uh, kind of rolls down.

:

01:07:53,423 --> 01:07:57,443

We really see this as, you know, if we do

our job right this year and we execute.

:

01:07:57,983 --> 01:08:01,073

We really see this as

scaling in:

:

01:08:01,673 --> 01:08:05,093

Um, you know, we'll end up with maybe

5,000 seats by the end of the year, but

:

01:08:05,093 --> 01:08:09,593

we'd like to get to, you know, 75,000

seats by the end of the year after that.

:

01:08:09,593 --> 01:08:12,623

So there's, there's, there's a plan to

kind of move quickly once it all works.

:

01:08:12,623 --> 01:08:17,183

So we're still in this phase of

setting up for success first and

:

01:08:17,183 --> 01:08:20,843

then rolling it out with scale once

we get into that kind of position.

:

01:08:21,288 --> 01:08:21,587

Randall Stevens: Great.

:

01:08:21,875 --> 01:08:22,205

Well, this has

:

01:08:22,205 --> 01:08:22,835

been fun.

:

01:08:22,925 --> 01:08:25,020

I could talk about this

stuff all day long and, uh.

:

01:08:25,537 --> 01:08:26,617

thanks for coming on.

:

01:08:26,617 --> 01:08:28,777

And, and Presanta thanks for sharing.

:

01:08:28,777 --> 01:08:33,037

You know, you've got, you've got,

uh, deeper technical, uh, you know,

:

01:08:33,037 --> 01:08:35,977

understanding 'cause you've been doing

this for so long and I'm, I'm sure

:

01:08:35,977 --> 01:08:39,397

there's been exciting last couple of

years just seeing, you know, the, the,

:

01:08:39,877 --> 01:08:44,197

what the last five years really with the

LLM development has opened up, right?

:

01:08:44,197 --> 01:08:48,487

All these possibilities of, of things

that you've been, uh, you know, thinking

:

01:08:48,487 --> 01:08:50,197

about, uh, over all these years.

:

01:08:50,197 --> 01:08:54,277

It's just unlocked all kinds of,

uh, all kinds of possibilities

:

01:08:54,277 --> 01:08:55,747

and value, I think, in the market.

:

01:08:55,747 --> 01:08:57,577

So I'm glad to see you guys working on it.

:

01:08:57,577 --> 01:09:02,527

And, uh, we, uh, uh, I mentioned that

we're doing these Confluence events, one

:

01:09:02,527 --> 01:09:06,577

day events, and it just so happens that,

uh, this year we're gonna be in Seattle,

:

01:09:06,577 --> 01:09:08,377

which is where you live, and we're gonna

:

01:09:08,377 --> 01:09:11,167

be in, in, Chicago in May, Michael.

:

01:09:11,167 --> 01:09:12,846

So, uh, looking forward to, uh,

:

01:09:12,877 --> 01:09:15,397

connecting with you guys at those events.

:

01:09:15,402 --> 01:09:15,442

Um.

:

01:09:15,714 --> 01:09:16,493

Any, uh, Any,

:

01:09:16,493 --> 01:09:17,904

last comments or thoughts Evan?

:

01:09:18,202 --> 01:09:20,362

Evan Troxel: No, it's been a

great conversation and, and I,

:

01:09:20,362 --> 01:09:23,211

you know, the, the thing that

I, I'm beginning to kind of.

:

01:09:23,272 --> 01:09:27,202

I try to wrap my head around is

the whole idea of kind of spatial

:

01:09:27,202 --> 01:09:29,992

awareness linked with ai, right?

:

01:09:29,992 --> 01:09:34,612

Because large language models

being language based and then kind

:

01:09:34,612 --> 01:09:36,261

of this link to the real world.

:

01:09:36,292 --> 01:09:41,452

I mean, I'm truly fascinated by the

challenge that you guys are undertaking,

:

01:09:42,022 --> 01:09:48,322

and I think that this really is kind of,

I mean, that's gonna be a big leap when

:

01:09:48,322 --> 01:09:53,692

that happens in, in AI and technology

in general, is, is that kind of spatial

:

01:09:53,692 --> 01:09:58,672

reasoning, spatial understanding, um, and,

and the environments that things sit in.

:

01:09:58,672 --> 01:10:03,592

Because I think as architects, a

lot of times we think about it as

:

01:10:03,592 --> 01:10:08,182

the project ends at the property

line, but in a campus situation.

:

01:10:08,987 --> 01:10:12,262

Um, I mean, that property line

is just an imaginary line.

:

01:10:12,262 --> 01:10:12,502

I mean,

:

01:10:12,502 --> 01:10:12,862

property

:

01:10:12,886 --> 01:10:13,306

Randall Stevens: You always have

:

01:10:13,522 --> 01:10:14,002

Evan Troxel: lines right?

:

01:10:14,002 --> 01:10:19,042

But, but, but there's, there's vegetation,

there's trees, there's other buildings.

:

01:10:19,042 --> 01:10:22,462

They have, you know, stacks on 'em

that are blowing out stuff from the

:

01:10:22,462 --> 01:10:27,262

chemistry lab and all of these things,

and all of that matters when it comes

:

01:10:27,262 --> 01:10:34,042

to the kind of analysis and reasoning

and, you know, decision making

:

01:10:34,042 --> 01:10:35,902

process that goes into digital twins.

:

01:10:35,902 --> 01:10:39,682

It's much larger than maybe,

a little building on an island

:

01:10:39,682 --> 01:10:40,792

in the middle of a lake, right?

:

01:10:40,792 --> 01:10:45,427

So, so there very different context

for every building and all of that

:

01:10:45,592 --> 01:10:50,332

kind of just starts to elaborate

on like the complexity that exists

:

01:10:50,332 --> 01:10:54,202

in the built environment and, and

how this kind of technology maybe

:

01:10:54,202 --> 01:10:56,872

fits into that overarching problem.

:

01:10:56,872 --> 01:10:57,292

So,

:

01:10:57,652 --> 01:10:57,653

Dr. Prasanta Bose:

:

01:10:57,653 --> 01:11:01,552

there is a, if you look at the, uh, one

school of thought that is now evolving,

:

01:11:01,582 --> 01:11:04,612

this notion of regenerative design, right?

:

01:11:04,612 --> 01:11:09,172

Regenerative design thinking and, and

the foundations that we are creating

:

01:11:09,412 --> 01:11:11,602

is actually rightly suited for that.

:

01:11:11,632 --> 01:11:15,112

When I said predictive is, I

basically, I'm looking at that as

:

01:11:15,112 --> 01:11:17,302

from a regenerative standpoint, right?

:

01:11:17,452 --> 01:11:21,262

I'm thinking, how can it

contribute in a better way, right?

:

01:11:21,352 --> 01:11:24,112

In a harmonious way in that ecosystem.

:

01:11:24,742 --> 01:11:29,392

So that's like a, and if you don't

think it right from get go, you can,

:

01:11:29,392 --> 01:11:30,622

you're not gonna be doing a patchwork.

:

01:11:30,622 --> 01:11:31,582

You cannot happen that.

:

01:11:31,657 --> 01:11:31,777

Okay.

:

01:11:32,548 --> 01:11:35,848

Randall Stevens: Well, thanks again

for joining us and uh, looking forward

:

01:11:35,848 --> 01:11:37,948

to seeing you guys when we're on the road

:

01:11:37,948 --> 01:11:38,428

in uh,

:

01:11:38,663 --> 01:11:38,664

Dr. Prasanta Bose:

:

01:11:38,664 --> 01:11:39,143

Oh yeah.

:

01:11:39,178 --> 01:11:42,803

I'd love to be, I'd love to meet you

and uh, have a great conversation.

:

01:11:42,803 --> 01:11:43,643

One more time.

:

01:11:43,877 --> 01:11:44,927

Randall Stevens: Alright,

thanks gentlemen.

:

01:11:45,112 --> 01:11:45,967

Michael Jansen: Thank you too.

:

01:11:45,967 --> 01:11:46,507

Appreciate it.

Next Episode All Episodes Previous Episode

Listen for free

Show artwork for Confluence

About the Podcast

Confluence
The director's commentary track for AEC industry software development.
The Confluence podcast is the director's commentary track for AEC industry software. Go behind the scenes with us to learn how and why decisions were made in the creation of your favorite software for the architecture, engineering, and construction industries.

It's a collaboration between Randall Stevens of AVAIL and Evan Troxel of TRXL.

About your hosts

Evan Troxel

Profile picture for Evan Troxel
An industry-leading design and technology expert with a passion for connecting people, Evan is a licensed architect in California and is most well known for his podcasts that focus on the AEC industry.

He has over 25 years of experience in the practice and technology in the architectural profession working with large teams to deliver large public projects for clients. He now brings his experiences together on the Archispeak and TRXL podcasts, and now on the Confluence podcast.

Randall Stevens

Profile picture for Randall Stevens
An AEC industry veteran with 25 years of software development, and sales and management experience, Randall offers a unique combination of expertise in software and graphics technology — coupled with a background and degree in architecture.

In 1991 he founded ArchVision, a software firm specializing in 3D graphics, specifically Rich Photorealistic Content (RPC). Through ArchVision, Randall has built an extensive network with the industry’s leading experts, architectural firms, and visualization software companies, which led him to product development of the AVAIL platform.