Christopher Parsons of Knowledge Architecture joins the show to talk about In this episode of the Confluence podcast, Chris Parsons of Knowledge Architecture joins us to discuss bringing AI to their Synthesis intranet platform. He takes us behind the scenes showing various functionalities including activity streams, document libraries, and project directories, with a focus on the upcoming integration of AI into Synthesis’ search and video transcriptions.

Chris also shows advancements in next-generation search, vector search, and future LMS capabilities. We also explore AI's impact on firm-wide knowledge capture, improved search relevance, and the creation of AEC-specific models through what Knowledge Architecture calls “Community AI”.

Episode Links:

• Christopher Parsons on LinkedIn
• Knowledge Architecture website
• Knowledge Architecture on LinkedIn
• Knowledge Architecture on Twitter
• Confluence podcast S1E7 - Overhauling Knowledge Architecture’s Synthesis v6 with Chris Parsons (YouTube)
• Synthesis roadmap - Bringing AI to Synthesis
• Synthesis intranet platform
• KA Connect Talk Archive
• Perplexity AI
• Deep Smarts: How to Cultivate and Transfer Enduring Business Wisdom by Dorothy Leonard (Amazon)
• The 3-Click Rule for Navigation Is False (full article)
• Stop Counting Clicks: The 3 Click Rule is Nonsense (short YouTube video)

Watch this episode on YouTube or Spotify.

-----

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

• Learn more about them at https://trxl.co/media and https://getavail.com
• Learn more about the Confluence podcast at https://confluence.getavail.com/podcast
• Learn about the Confluence events at https://confluence.getavail.com

Welcome back to the Confluence podcast.

2

we have Chris Parsons joining us again.

3

Hopefully you've tuned in and

heard the first episode that we

4

did with Chris a few weeks ago.

5

But if not, you might want to

go back and listen to that.

6

But, uh, happy to have

you on again, Chris.

7

Uh, you know, I know we're going

to, we're going to dig more

8

into the Synthesis platform.

9

And really start to dig into, you know,

not only what, what Synthesis does, which

10

I'll get you to kind of kick this off

with, but ultimately like, why, you know,

11

how AI is affecting what you're doing

and how you guys are thinking about that.

12

So, uh, maybe you can just give us a

good overview, introduce yourself in

13

a good overview of what Synthesis is.

14

Christopher Parsons: sure.

15

so for folks who didn't hear their

16

We are a hundred percent focused on

AEC Synthesis is an intranet platform

17

for AEC firms, and we work with about

130 firms, mostly us, although we

18

have some international, and we are

deeply integrated with a lot of AEC

19

software that people care about.

20

So Deltek, Uninet, OpenAsset,

Newforma, AEC 360, Zendesk, and so on.

21

So it's a, it's Synthesis is, I think,

pretty well named product in that it

22

like connects data from multiple places,

giving you one kind of searchable

23

source of central source of truth.

24

And, um, I am just going to do, I'm

going to do like a high level seven

25

screen kind of like look at Synthesis.

26

So people get like a, a ball, a kind

of general idea of the kind of content

27

that we have so that when we start

talking about what we're doing with

28

AI, it'll, it'll make a lot more sense.

29

So I'm going to go ahead

and share my web browser.

30

So this is the homepage of Synthesis.

31

This is a demo environment.

32

So obviously you can brand it.

33

Um, colors and fonts and logos and all

the things to make it feel like home.

34

But this will give you a general

sense of what the product is.

35

So most of our clients on the homepage

will have, um, an activity stream.

36

So this will be posts and comments around

different things happening in the company.

37

So it's very much an internal

communications platform.

38

So whether it's a strategic plan update

or a project tour or a new piece of

39

software has been added or some PR.

40

you know, somebody's retiring,

you know, these are the kinds

41

of things that we typically will

see on a Synthesis intranet.

42

And if I jump into one of those

posts, um, you'll kind of see that

43

it's, it's a very rich multimedia.

44

You can have images and videos and

links, and then we've got comments

45

and likes and the kind of things you'd

expect to see, um, on a social intranet.

46

Um, if I go across the top, um, we've

got our mega menu navigation, and this

47

is kind of just showing you the breadth

and depth of content that typically

48

will end up in a Synthesis intranet.

49

So kind of.

50

Um, on home, you'll usually find like

whelping to the company and stuff like

51

mission values, the leadership team, the

different committees, um, different office

52

stuff, um, in an HR community that we

call these communities across the top.

53

So you'd find HR information, professional

development information, time and travel,

54

accounting, different office pages.

55

Projects in practice, things like

standards and codes, um, project

56

management resources, different

directories to projects, um, in

57

marketing, kind of brand templates,

marketing resources, technology, the

58

different platforms that people use

and the technology policies, and then

59

learning, you know, different, you know,

whether it's a company university or

60

different things like software training.

61

So this is what we call a

video library in Synthesis.

62

Um, These have got different training

courses, depending on different

63

platforms that the company's using.

64

Um, video libraries have filters, so

you can add custom fields and kind

65

of drill into those different videos.

66

Um, and it's all searchable.

67

So I'm, I'm going to skim across the top.

68

So video is going to become important.

69

We're going to come back to this

later and we'll come back to

70

this idea of a video library for

training a little bit later as well.

71

Um, I want to talk about another

key piece of content and Synthesis,

72

which are document libraries.

73

So if I go into our projects and practice

community and go to our standards library.

74

You'll just see that we've got

some different standards here

75

organized by CSI division.

76

And again, you can add custom fields

into Synthesis so it can be on

77

CSI division or standard type or

conditions or whatever the different

78

things you want to apply are.

79

So that's videos and documents.

80

Um, from a page perspective, I'm

just going to go into technology

81

and I'll go under our design

technology section and go to Revit.

82

And this is just a landing page in

the company for all things Revit.

83

So it's got links off to

other different resources.

84

Um, we've got training videos on

Revit in the upper right, I've got

85

kind of our key points of contact.

86

So I've got, you know, Brianne and

Julie, and then I can kind of reach out.

87

I can see their team status, et cetera.

88

And then I can see any recent posts

that were written about Revit.

89

So this is a pretty typical, like, so if

you imagine this kind of layout, but for

90

a lot of things within the company, you're

trying to connect people to the right

91

people and the right resources and helping

them know how the company approaches, you

92

know, whether it's a piece of technology,

whether it's a policy, um, et cetera.

93

If I go into projects and practice,

a big, uh, kind of content type

94

for us is what we call guides.

95

So I'll go to the

project management guide.

96

And so this technology basically

replaces like the hundred page PDFs

97

that your firm might use for your

employee handbook, or in this case,

98

your project management manual, or a

CAD or BIM guide, or a brand guide.

99

There's all these different like kind

of guides that we saw at our clients.

100

And so we built a tool that

handles that content really well.

101

So you've got kind of the different

people in project management.

102

If I kind of go into contracts.

103

You know, I can, you've seen, we've

broken what used to be a hundred page

104

PDF is into a bunch of small chunks.

105

And then these are all kind of like,

you know, I've got a table of content

106

and I can kind of jump through our

different approach to contracts or

107

billing rates or how we start projects or

financial controls, like all the things.

108

Um, this is a nice place to

put it within the internet.

109

Um, so three more things

I want to show you.

110

Um, if I go to our project

directory, so this is, I mentioned

111

some integrations that we have.

112

This is where those integrations with.

113

You know, Newforma, et cetera, come in.

114

It helps us pull information from

those systems into one place.

115

So I can do things like find all of our

projects by type, you know, or by state.

116

Um, this is all searchable as well.

117

Um, or I can put them on a map, right?

118

And I can drill in and say like, well,

what's the work we've done in California.

119

And I can see we've got 11 projects

in the Bay area, and then two of

120

them are out in Marin, et cetera.

121

So I've got a good way visually

to kind of explore the work,

122

um, that the company has done.

123

Um, and if I click on one of these

projects, I get a project profile.

124

So this is data we're pulling

in from all those systems.

125

So I've got imagery from OpenAsset,

project type, cost and size, location,

126

descriptions, like a lot of rich data,

as well as who worked on the project,

127

both inside and outside the company.

128

If I click on one of these employees,

I get to an employee profile.

129

So this is our chief

executive officer, Em Davison.

130

And again, I'm pulling information

in from multiple systems.

131

We've also got custom fields

and Synthesis you can use.

132

And you can see things like professional

bio and education and skills and her

133

posts and what projects she worked on.

134

And then if I go all the way back up

to the employee directory, it's that

135

same kind of idea with projects, right?

136

I can filter employees on a

variety of different fields.

137

Um, and search for them that way.

138

And the last piece I want to show

you, because this is very pertinent

139

to what we're doing today, is search.

140

So I'm going to search

on something like Revit.

141

You can see that I've got some

suggested search results, but if

142

I hit return, I see everything.

143

And so I see what we call best bets.

144

So Rob Thompson's kind of

our go to person for Revit.

145

But if I keep scrolling down, I see,

I'm, we're searching across pages and

146

posts and documents and projects and

videos and all the stuff in Synthesis.

147

So.

148

So that's kind of Synthesis like at

a high level, um, the kind of main

149

moving parts that make up the intranet.

150

Evan Troxel: And Chris, the last time

that you were here, you showed us this.

151

Really great graphic that you've come

up with this periodic table where

152

you kind of talk about the different,

you, you can probably say it better

153

Christopher Parsons: Sure.

154

Evan Troxel: you have these categories

with different topics within each

155

category that are kind of knowledge

156

Christopher Parsons: Yeah.

157

Yep.

158

Evan Troxel: and that kind of shows that.

159

There's a lot going on and you

just showed like how all of that

160

manifests in an intranet but like

there's people responsible to get

161

that content into the intranet.

162

I always found that was one of the hardest

things to do was to get people to actually

163

go do those things on the intranet to make

it available for everybody else, right?

164

Um, it is a step way above

and I know Randall, you have

165

a special place in your heart.

166

from a content management standpoint

as well, of like, just making sure

167

that the data is in there and that

people can find it and it's accessible,

168

but it still relies, you know,

this is way better than files and

169

folders on a, on a server somewhere.

170

There's graphical aspect, there's layout,

there's, you can do all kinds of things.

171

You can link between pages

much more easily than you can

172

do between files on servers.

173

You're still relying on people.

174

And I think this is where we,

we, we start to talk about the

175

future of Synthesis as well.

176

Right?

177

Which is.

178

Like the elephant in the room with AI,

but up to this point, we've relied on

179

people to do all of that work you're

showing this beautifully idealized version

180

Christopher Parsons: Yeah.

181

Evan Troxel: And then, and then I think

about like the reality of someone's

182

SharePoint intranet or whatever they've

got going on and like half the data is

183

on the right page and that person who

did it no longer works at the company

184

and there's all kinds of other stuff.

185

And I just want to talk about,

I just want to throw all of that

186

out there because these are the

realities of working in a firm where.

187

It's like, this is another job.

188

Like we need content managers now to,

to make sure that this stuff stays up

189

to date, that people are keeping an

eye on the community, that it actually

190

is grease between the gears of the

firm, you know, and all these different

191

departments and how they interconnect.

192

And a lot going on there, but this is

potentially where AI and things like

193

that can actually help because, you're

going to talk about some more specific

194

examples, but you can throw data at it

and have it pull insights out of that

195

in a much, I don't know, easier way.

196

Let's just call it easier than it would

be for someone else to comb through a

197

document, to split it up into all those

different topical, you know, the table

198

of contents that you showed in the,

in the standards area, there's video.

199

That, you know, has been typically if

somebody posts an hour long video, no one

200

wants, wants to watch an hour long video.

201

How can I find what

202

Christopher Parsons: Right.

203

Right.

204

Evan Troxel: there's a lot of

challenges still in the content

205

that even gets posted, let alone

just the management of all that

206

Christopher Parsons: I think you're right.

207

And I think we will dive into that.

208

I think just at a high level, a

couple of ways I think about what

209

you're saying are, one, making the

software easier to use has been a huge.

210

level, easier to use and

then more beautiful the end

211

product that you make with it.

212

Right.

213

That's, that's been really important,

especially we're a hundred percent

214

AEC and that matters to the designers

and, and engineers in our community.

215

Um, I think the second piece is raising

the ROI of having that knowledge captured

216

in the platform in the first place.

217

And so when you see what I'm about to

show you with where we're going with

218

search and AI, I think people start

getting like, and this has happened

219

already in our journey with Synthesis.

220

It's like, oh, I get it.

221

So if we put our data in Deltek,

that means I can get it out in

222

Synthesis in this beautiful way.

223

Or the same thing with

images from open assets.

224

So by connecting systems, you raise the

return on investment of that information.

225

Um, and so when you start seeing what

we're going to do with AI, I think

226

it's going to create kind of like a

gravitational pull to get more content

227

into the system because you'll realize,

Oh, if we have our best content, we're

228

going to be able to do, and I know Randall

and the folks at Avail are investing

229

in this as well, you know, the better

content that I get into my system,

230

the further we can go as a company.

231

And I think that's just going to be a

step change from where we are today.

232

Randall Stevens: I was, I was going

to make a comment, Chris, that I,

233

you know, I think you already Made

note that Synthesis is a great name.

234

You know, those names usually

come about after you kind of

235

figure out what you're doing.

236

It's like, this could only be named

this if we, if we do it properly.

237

But, um, you know, I've just made

a note that, you know, I think

238

a challenge of these systems.

239

of adding things like a Synthesis or

avail, uh, into an operation is to start

240

trying to figure out how not to be yet

another thing that somebody has to do.

241

But, but the by product of, of the main

things that you're doing and the workflows

242

that you're doing should as automatically

as possible flow into these things.

243

And, you know, I think about, you know,

when you show a Synthesis and I don't

244

see how if you're a large firm that you

don't have, you That as your, you know,

245

it's, I think of it as like, that's your

newspaper for the, that's your, that's

246

your news source to keep everybody

contextually aware of what is going on.

247

And as a firm scales, there's just no

way that everybody can kind of understand

248

what everybody in the firm is doing.

249

So it's a very, to me,

it's a very elegant way.

250

And then the trick, like you said,

is it can't become, Somebody's job

251

to go do all of those things because

it falls apart pretty quickly.

252

So it's gotta be a, a by

product of the normal workflows.

253

And I know that's what we were

always kind of, uh, trying to do

254

is to figure out, uh, I don't want

to give you another job to do.

255

I want to, I want you to

go do what you normally do.

256

And then we'll try to, you

know, be a by product of that.

257

Christopher Parsons: Totally.

258

I do think though, like, yeah, as

a good example, like I showed that

259

project management guide, like the

folks that put those things together.

260

And I know you both work with firms that

have probably done things like that.

261

Like those are labors of

262

love done by a couple of key

individuals in the organization.

263

Um, I don't think they were

generally happy about making

264

hundred page long PDFs with them

because they're very hard to edit.

265

Then you have to redistribute them.

266

Like it's a terrible, like authoring

platform for something like that.

267

So,

268

and it's a dead document.

269

Yeah,

270

Evan Troxel: now I have to keep this

271

Christopher Parsons: correct.

272

Yep.

273

Evan Troxel: the time.

274

And so like that, the more Wikipedia

style knowledge base that is

275

constantly updated with multiple

276

Christopher Parsons: Right,

277

Evan Troxel: who can do some

accountability with each other and

278

they can hand it off and make, you

know, if they don't have, they're

279

working on a project, somebody else can

come in and do it makes so much more

280

Christopher Parsons: exactly.

281

Evan Troxel: Just to keep

the life in these documents

282

because they are constantly

283

Christopher Parsons: you can put

short training videos in the middle

284

of these, which is harder to do

with a document and you can add

285

links to people and other documents.

286

So yeah, it's a much better

experience, but to Randall's point,

287

like this isn't a, somebody's job

288

is doing this.

289

Like, and so hopefully we're

giving them a better way to do it.

290

Um, I, so that's kind of,

if I can share my screen,

291

Evan Troxel: Mm-Hmm.

292

Christopher Parsons: Um, so

I, I have shown you kind of

293

where Synthesis is today.

294

And what I want to be talking

about is where we're kind of going.

295

And so our mission, like kind of

the way we'll describe the product

296

in the future is around being an

integrated intranet LMS and enterprise

297

search platform for AEC firms.

298

So I touched on the intranet piece.

299

I want to touch on Synthesis LMS.

300

I've just got three slides on it because

again, it'll get a sense of how the

301

intranet LMS and enterprise search will

all come together and the kind of content

302

that will live in Synthesis in the future

and kind of like what we're building for.

303

Um, so this is, and this is a mock, right?

304

We're in design now.

305

We're going to start construction

this summer on this project, but

306

this is a course in Synthesis LMS

called Project Communications.

307

And I'm in a lesson called

Client Communication Strategies.

308

And in this example, there's six

lessons on the right hand side.

309

You can see that I'm through one lesson.

310

Um, I can see who the

instructors are on the right.

311

And it's a, you know, an 18 minute

video plus a handout, right?

312

On active listening fundamentals.

313

So we have seen our clients

over the years, try and use the

314

intranet platform to do this kind

of stuff and do reasonably well.

315

And we kind of call it LMS Lite.

316

And we had just come to the point

after finishing Synthesis 6 where we

317

went out to our clients and had a big

listening tour and they're like, We want

318

you to work on two things, AI and LMS.

319

So this is the LMS piece and we're

going to talk about the AI piece.

320

Um, all those courses can get

bundled up into a course catalog.

321

So in the example I'm showing you here,

there's courses around onboarding and

322

project management, but you know, our

clients create courses around things

323

like design technology, sustainability,

leadership, design itself, marketing

324

and business development, and so on.

325

So this is meant to be one central place

people can have all of their educational

326

Randall Stevens: Chris, do you, do you,

327

are you hosting those videos or do you

support Vimeo and YouTube and other places

328

that are, you know, traditional places

that people would host video content?

329

Christopher Parsons: Uh, yes to both.

330

So we have a feature called

Synthesis Native Video.

331

We, uh, released in 2018.

332

We are in the process of

doing a major overhaul.

333

We'll actually talk about that toward

the end of the conversation around us

334

using AI to generate, um, AEC specific

transcriptions for those videos.

335

We're also adding things like

chapters and some other neat

336

features, but no, the Synthesis

Native Video is a key part of that.

337

of kind of how all of this

338

comes together.

339

And we're going to talk

about that in some detail.

340

But we also do, you know, through embed

341

codes, you know, you can drop in

Vimeo, YouTube, that kind of thing.

342

and the last piece about the

Learning Center is the ability

343

to bundle those courses up into

what we call Learning Paths.

344

So for example, if you look

under onboarding, there's two

345

learning paths on onboarding.

346

There's one that's called

general onboarding.

347

That's got like seven courses.

348

And then there's another one

for project manager onboarding,

349

which got, has got 10 courses.

350

So you take that kind of core seven

courses for, you know, everyone

351

that comes and joins the firm.

352

And then you add three more to

like onboard project managers.

353

So the way we do project

management here at the company.

354

So, you know, it will also have

assignments and transcriptions and

355

analytics and all the kind of basic

things you'd expect to find in an LMS.

356

And this is scheduled to go

357

So we have a 30 minute concept

video laying all this out on the

358

roadmap, page on our website.

359

Maybe we can drop it in the show

notes, but that's as far as I want to

360

take it for what we're doing today.

361

So let's talk about AI.

362

So for us, it's three things,

um, that all work together.

363

There's next generation

search, which we'll start with.

364

There's video captions, which Randall,

you just kind of teed us up for.

365

And then we have a program that we

introduced called Community AI, which

366

is going to help us build AEC specific

models to make this work even better.

367

Um, so now that we've kind of

seen the content, I want to do

368

a deep dive into how we're going

to make it even more searchable.

369

So as we saw in that quick search

example, today we're using keyword search.

370

Um, we've, we've done pretty well

with keyword search, but we've taken

371

it about as far as we think we can.

372

So I'm searching on keyword like jury

duty, and maybe what I really wanted

373

to do was ask a question, right?

374

How many hours does our company

provide for jury duty service?

375

Or what's our jury duty policy?

376

And when I execute that search, what I

get back are links, in this example, to

377

pages and to documents, which, you know,

that's how we've generally done search.

378

Um, where we're going is, you know, using

generative AI and other kind of advances

379

in search technology to make answers.

380

That summarize useful information from

multiple different sources automatically.

381

So for example, for this question

about what's our jury duty policy,

382

we're giving part of an answer

around allocation of hours.

383

So how many hours do

we give documentation?

384

So how do you kind of like

prove you were on jury duty?

385

And then how do I request time off the

ability to ask follow up questions?

386

And then if I drill in to one of

these, um, search summaries, I can

387

link to that source, you know, that

we, and we're citing the source

388

where we got that information.

389

So another example, I'm

searching on worksharing.

390

And what I really want to ask is, should

I use Revit worksharing or central files?

391

And then I want to get, you

know, an answer back, right?

392

So it's summarizing the benefits

of worksharing, the benefits

393

of central files, and then kind

of giving a recommendation.

394

Um, on the right, you can see the

top sources that were used in kind

395

of generating that search summary.

396

And again, I can drill in and find

out the underlying information that

397

was used to generate this answer.

398

And if I click on it That will take

me directly into that video and

399

directly into the part of a video,

um, that contains the passage that

400

was used to generate that answer.

401

So this is the Synthesis native video,

402

and I'll throw it to

your earlier question.

403

That's how we're making this work.

404

Evan Troxel: So that's interesting

because you're doing things that I think

405

YouTube is trying to do as well, right?

406

Where, where they're automatically kind

of looking, they're transcribing, they're

407

looking at the conversation there.

408

I know whenever I upload a YouTube video,

it's like, do you want to automatically

409

it find moments or whatever they call it?

410

They don't even call it chapters.

411

Like you can put in your own

chapters in the description field

412

to help people navigate to a certain

thing that I think you might want

413

to navigate to, but they're also.

414

looking at finding smarter ways to

find those key moments in videos.

415

But then you're, you're locking kind of

a timestamp and what was said to a, to

416

a question, which makes it much easier

to, like I was talking about earlier,

417

no one wants to watch the hour long

video too long, didn't watch, right?

418

I just want the key takeaways,

or I want to jump to a specific

419

point in that video that is

regarding the question that I have.

420

So that's what you guys are really

taking on here with your own

421

Christopher Parsons: Right.

422

Evan Troxel: You can then process that

in the background and, and then apply

423

all of that metadata back into the, the

424

Christopher Parsons: Correct.

425

And I think that the other reason this

is a huge unlock is, you know, because

426

we work 100 percent AEC, and I'm going

to touch on this a little bit later.

427

There's so much AEC terminology

and lexicon acronyms and product

428

names and material names and all the

stuff that we do in this industry.

429

So much lexicon stuff.

430

So,

431

Evan Troxel: That's where YouTube

432

Christopher Parsons: right, exactly.

433

Evan Troxel: Because it, it says Rabbit

434

Christopher Parsons: You got it.

435

Evan Troxel: and and there's all,

you know, that's just one example,

436

but there's so many acronyms.

437

There's so much terminology and jargon

and vocabulary that, you know, it's,

438

it's great when you're in an office,

you need to be using that because

439

you're all working in the same field

and you're all doing the same things.

440

Thanks.

441

And it, you have to be able to then, like

you're saying, like, it's a huge advantage

442

to, have that vocabulary be a part of the

443

Christopher Parsons: Exactly.

444

Evan Troxel: that it finds

exactly what you're looking for.

445

And it's not it as a misinterpretation.

446

And now I can't even find it

because I didn't search for rabbit.

447

I searched for Revit,

448

Christopher Parsons: right?

449

Or I searched on standards

and we use guideline.

450

Like there's so much stuff

451

like that.

452

And so, um, you're teaming up

really, really well, actually.

453

So that's it for, I've got

for like of the front end.

454

And this project I've been talking about

as an iceberg all along, hopefully not

455

in the Titanic sense of the word, but

in the 10 percent of this project is

456

UI above the waterline, 90 percent of

what we're doing to bring that search

457

stuff to life is under the waterline.

458

It's invisible.

459

It's infrastructure behind the scenes.

460

And what I want to share with you and

your community is kind of what's below

461

the waterline and like how the generative

kind of search results actually work.

462

Yeah, please.

463

Totally.

464

Evan Troxel: grown up with the

internet, And so we've all been trained

465

to use the search tools that are on

the internet to our advantage, which

466

was typically keyword based, right?

467

So I'm a keyword searcher and it is

such a mental leap for me to ask a

468

computer a well formulated question.

469

It's like you go into the chat GPT, you

go into Claude, you go into whatever.

470

And, and, It's funny because my, my

wife will, she, she adjusted and adopted

471

this way of doing a search a while

ago, which is just ask it a question

472

and actually put a question mark at

the end of it in the Google search.

473

I still don't, I still can't even

do, I can't bring myself to do that

474

because it's not my muscle memory.

475

And so I'm wondering like

with firms and, and how you're

476

creating an integration with that.

477

Cause I, I would assume that.

478

If I don't even know the right question to

ask, but I just know, okay, work sharing

479

Christopher Parsons: Right.

480

Evan Troxel: I can type in work

sharing and then maybe there's Maybe

481

there's even other questions that

have been, that have come up from

482

other people about work sharing.

483

So it's like, Oh yeah, that's, that's

really what I want to ask because

484

me, I'm, I'm just a work sharing.

485

I would say Revit work sharing, right?

486

Christopher Parsons: Totally.

487

Yeah.

488

Evan Troxel: see what comes back.

489

I don't ask it the question.

490

And now this is all, this is

very much a communication.

491

This is a conversation with,

with the computer at this point.

492

And I'm not conversational with computers.

493

That's just not how I'm

494

Christopher Parsons: Yeah, I think you're,

you're on to, and I think you're right

495

for probably representing most users.

496

Um, and so I think for us, we have to

do really well when people are still

497

using keywords in Synthesis search.

498

And I think, you know, for example, let's

take something really prosaic, like Wi Fi.

499

Like if somebody's asking, if someone

types in Wi Fi, what are they asking?

500

They're not asking which

Wi Fi technology do we use?

501

Not asking, like, do we have Wi Fi?

502

They're asking what's

503

the, you know, the, Wi Fi network in

my office and what's the password.

504

That's what they want to know.

505

Right?

506

So.

507

Randall Stevens: Every

508

firm needs that.

509

Every conference room I go into,

they cannot get me on the Wi Fi.

510

I

511

Christopher Parsons: from a meeting

people where they are perspective,

512

but also from just an efficiency, like

why type 10 words if I can, should be

513

able to get what I need with one word.

514

Um, and you know, the more

people type, the more the

515

typos spike, you know, all the

516

things, um,

517

yeah,

518

Randall Stevens: you know, you can,

you can begin as you're capturing, uh,

519

You can begin to make those suggestions

as somebody starts to type something in.

520

It's like, Hey, these are the

popular things that have been

521

asked before, even for the user

to not have to retype it, right?

522

It's like, okay,

523

Evan Troxel: So useful, right?

524

It's like auto complete for

what I was thinking, not even

525

Randall Stevens: Now, you know, Google's,

uh, Google's gotten really good, right?

526

It's like content, you know,

it's contextual awareness.

527

Uh, and as you start to type

something, it's like, it knows

528

what I'm getting ready to,

529

Evan Troxel: They've got a

530

Randall Stevens: Right.

531

Christopher Parsons: They

got a pretty big user base.

532

Yeah.

533

I mean, I think that's one,

the, the tool that I've been

534

super inspired by is Perplexity.

535

I don't know if either of you have

used that, um, yet, but it's fantastic.

536

And it like really handles that.

537

from one word all the way

through a complete, you know,

538

multi sentence question.

539

Um, well, and so it really depends.

540

And sometimes like one word does it,

or two words do it, whatever it works

541

sharing would be fine, or if it's like,

yeah, I see what you answered, but like,

542

really now I know what it is I'm asking.

543

What I'm asking is, you know, X, Y, Z.

544

So.

545

Evan Troxel: That's usually

where it leads me, right?

546

It's like, okay, now, okay, now

that I've seen a little bit more,

547

I can ask a little bit better of a

question, and like five questions from

548

Christopher Parsons: That's right.

549

100%.

550

Evan Troxel: wanted.

551

Christopher Parsons: Yeah.

552

Back to the iceberg.

553

So, um, when we, when we started

approaching, you know, obviously

554

when everything changed with

chat GPT, um, we started.

555

Asking the question, like every tech

company, well, every company writ large,

556

not just tech companies we're asking.

557

It's like, okay, what

does this mean for us?

558

What does that mean for our clients?

559

And we quickly kind of keyed in on search

and video transcription as being too

560

important, uh, things we wanted to build.

561

They weren't little quick hits, you

know, we didn't rush something out.

562

Um, but we really took a lot of

time thinking about architecture.

563

And so one of the ways we could

have built this was to build an end

564

user facing large language model.

565

Right.

566

Um, and so questions

would come into Synthesis.

567

We'd use a LLM that we had trained

and we generate answers, right?

568

This is basically, it's overly simplified,

but this is basically what ChatGPT does.

569

And for a host of reasons, like this

was the wrong architecture for us.

570

Um, the challenges are that

it is prone to hallucinations.

571

Just because of its data set, um,

well, oftentimes, you know, that goes

572

across, you know, different companies.

573

It can be a really, a really big problem.

574

Um, it's got a problem with freshness

because, you know, you train this model

575

and then you, you know, whatever, train

it nine months or 12 months later.

576

And so it starts getting a gap in terms

of stuff that's either you're holding on

577

to old information too long, or you're

not recognizing new stuff fast enough.

578

And then it's got a permission

issue, like inside a company.

579

It's really hard, like if we have,

for example, uh, uh, Principals or

580

a shareholders or a management team,

community and Synthesis, like building

581

an end user facing LLM, like who can

see what, what it's used to generate.

582

It's just too much.

583

It's too complicated.

584

You know, it's not, it's not

what we like the sound of.

585

So it's like, all right,

is there another way?

586

And what we came across, uh, is this

emerging approach called RAG, which

587

stands for retrieval augmented generation.

588

And we love it.

589

And this is kind of how

it works in a nutshell.

590

So a query comes in, like,

what's our GRDD policy?

591

We looked at this earlier.

592

We execute a search.

593

And so we execute a vector search.

594

And I'm going to come back

to that in a little bit.

595

Um, the vector search retrieves passages

of key text from those relevant resources.

596

So out of videos, the right passage out

of a document, you know, it's on page 20.

597

The right passage out of a post or a, uh.

598

You know, a page in the platform

sends those passages to the large

599

language model, which will then

summarize it, um, and generate those

600

answers and citations we looked at.

601

So we're not kind of indexing,

like all of the knowledge to make

602

an Oracle that knows everything.

603

We're using search and then we're just

using the LLM as a summarization tool,

604

cause it's very good at doing that.

605

Evan Troxel: Yeah, so is it, would

you call it more of a just in time

606

kind of a A machine, or, I mean,

607

Christopher Parsons: Yeah,

I think that's great.

608

I mean, I think I, if I put a post.

609

Up right now, it gets, by the time that

it's done indexing and it gets embedded.

610

We'll talk about that a little bit later.

611

Then the next time I do a search one

minute from now, and it's something

612

that was related to that post I just

added, that would be included in that

613

search and fed into the search summary.

614

So it very much is a,

615

Evan Troxel: this goes back to that dead

616

Christopher Parsons: exactly.

617

Yeah.

618

The freshness issue.

619

Evan Troxel: that you you brought up a

620

Christopher Parsons: Yeah.

621

So it was super important to get

something that really tackled, you

622

know, well, let's talk about it,

that really tackled hallucinations,

623

freshness, and permissions.

624

So.

625

You know, we counter hallucinations

by grounding the search in your data.

626

So this only searches, you

know, your company's data.

627

It doesn't do anything across,

you know, our different clients

628

or from the internet, right?

629

This is really your content.

630

Um, so again, that's that search that

retrieves the passages, passes it to the

631

LLM and then generates the citations.

632

Um, the real time search,

you know, you just mentioned.

633

That's a big part of it.

634

And then permission search.

635

So if Randall is a principal and Evan,

you know, you're a job captain, you're

636

going to get different, uh, passages

retrieved based on your permission

637

level that get fed into the LLM.

638

So we can do this summarization,

but we can do it in a way

639

that honors permissions.

640

Evan Troxel: So when people,

how are the permissions handled?

641

Are they handled during like the

upload of the new information

642

or is it based on places?

643

I'll generically call it places in the

644

Christopher Parsons: yeah,

645

Evan Troxel: So if it's like HR,

HR, everyone in HR has access to

646

everything in HR, but outside of that,

there's like certain eyes only, right?

647

So how do you handle that when new

information is constantly being uploaded?

648

Because.

649

being added to that.

650

Still going into kind of documents

and, and then like there's permission

651

for that document in that folder that

the LLM is in looking at when you ask

652

it a question or how do you handle

653

Christopher Parsons: yeah,

it's, it's very much the latter.

654

It's based on places.

655

We call them communities and

we have public communities and

656

private communities and Synthesis.

657

And so public communities

are what they sound like.

658

Anyone can see anything in that.

659

Within a public community, you have

different permissions for who can edit

660

content versus that are end users,

but everyone can see that content.

661

And in a private community, you have

to add, you know, users to it manually.

662

So that kind of, that's as far as we

took it, you know, like SharePoint,

663

as you two may know, like you get so

granular, their permissions, you're

664

down at the document level, it makes

things super complicated and people

665

can't remember who can see what.

666

So we.

667

Launched Synthesis 6 with a

very simple permission model.

668

And it's stood the last couple

of years of testing really well.

669

So that's, um, simplicity is

always good if you can get it.

670

So that's kind of the basics of

retrieval augmented generation and

671

you know, how it improves search.

672

So there's kind of three other

concepts that are important to how

673

we're building, uh, vector search,

which I'd lightly referenced earlier.

674

Um, Our community AI program, where

we're building those AEC specific

675

models and, uh, our, kind of our

approach to enterprise search.

676

Um, so vector search is super interesting

and, um, I'm going to overly simplify it.

677

And if there, I'm sure there are

some people in your audience and

678

your community who know about vector

search, and they're going to be like,

679

Dude, you're totally oversimplifying.

680

It's like, yes, I am acknowledging

I am oversimplifying this.

681

Um, so in,

682

yeah, so think of it as like a multi

dimensional space with thousands

683

of possible vectors, right?

684

Um, and so basically what we're doing

with vector search is we're turning, uh,

685

text into numbers and coordinates really.

686

So for example, programming might be

near floor plan or space planning.

687

in one sense of the word.

688

Programming may be near coding

or software development, kind

689

of in another sense of the word.

690

And programming might be near

scheduling or event planning in

691

yet another sense of the word.

692

Um, so let's go back to, uh,

to RAG to see this in action.

693

So when that query came in this

time around, do we have any

694

healthcare programming templates?

695

We execute a vector search.

696

which goes out to a vector database.

697

So it's stored all of the locations of all

of our content using those coordinates.

698

It then retrieves the passages of

the key text from the most relevant

699

resources, sends them to the large

language model for summarization,

700

and then it can do the citations.

701

So the question is, how does the vector

database know which passages to retrieve?

702

And the answer to that is getting into

our vector search ingestion pipeline.

703

Um, which is exciting as it sounds.

704

Um, so, so we, whenever anybody

uploads or edits a Synthesis resource,

705

uh, that resource gets broken

down into smaller chunks of text.

706

So for example, let's take a Synthesis

page and this is a simple one on time off.

707

So we're going to break that down

using the page headers into chunks.

708

So we've got a chunk around bereavement

time, a chunk around jury and

709

witness duty, and a chunk around PTO.

710

Um, or this is a feature in

Synthesis I didn't show, but these

711

are collapsible sections, right?

712

So we've got all these

different collapsible sections

713

having to do with payroll.

714

One on W2, one on how are

raises and bonuses decided, etc.

715

So we use those as signal on where

to break up a page, and then we

716

put them into those smaller chunks.

717

we take those chunks, and we send them

to what's called an embeddings model.

718

And I'm going to go into this in a little

bit of detail, but the embeddings model

719

knows how to look at those chunks of text,

And that it embeds coordinates, it embeds

720

vectors into them based on their meaning.

721

And they look like this.

722

It's really just a string of numbers.

723

It's complete nonsense, but

the, the vector database and the

724

embeddings model know what it means.

725

And so we add those chunks in the right

location using that vector, right?

726

And so this will come a little bit clearer

at the next step, but the key to this

727

whole organization is the embeddings

model, because it establishes the vector

728

space in which meaning is assigned.

729

Um, And because this is

such a key, am I good?

730

Should I pause or should I do, are there

any questions you guys want to ask?

731

Or should I keep, keep me going?

732

Yeah, please.

733

Yeah,

734

Evan Troxel: that I know we'll be

talking a little bit about, but

735

it's just a run on conversation.

736

And, and so there aren't these

Chunks that you're talking about

737

Christopher Parsons: right.

738

Evan Troxel: And I know, like, I think

this goes back to like the thing that

739

we were talking about earlier with

YouTube kind of trying to identify

740

Christopher Parsons: Yes.

741

Evan Troxel: And so maybe there's

sentiment changes, maybe there's.

742

There's pauses.

743

I don't know how it's doing it.

744

I mean, maybe you have

more insight into that.

745

So my, my first question is, is

there, it's like when, when the

746

document does not have any kind of

hierarchy to it, there's no structure.

747

This is what AI seems to

be pretty good at, right?

748

You can think of it.

749

Throw like a giant thing and it

can, it can kind of figure that out.

750

Like I use this all the time.

751

I use

752

Christopher Parsons: Yeah.

753

Yeah.

754

Evan Troxel: chat GPT too.

755

You can, you can summarize documents.

756

You can say, tell me the three key point,

key takeaways from this conversation.

757

And it does a pretty good job at that.

758

And so I'm just wondering, is that, is

that really helping you here as well?

759

And is that, is that really

how you're attacking this

760

Christopher Parsons: I think so.

761

Um, I, the reason I say, I think

so is we're evaluating a couple of

762

different approaches and we don't know,

today on April 25th, which one we're

763

going to pick, but that is certainly

one of them is to, and you know, it

764

doesn't have to be, it doesn't have

to be a hundred percent precise.

765

It just has to be good enough,

you know, to break it down.

766

Like this, we're going to be able to

find, you know, what we need anyway.

767

But, but I think your point is right.

768

That's, and that is, you can read,

I've actually read a lot about how

769

the Google key moments things works.

770

And it's a lot of what you're saying.

771

It's like looking for

changes in inflection.

772

It's looking for.

773

breaks it in some cases it will even

OCR the slides if there's slides or

774

the video and kind of get a sense

that something's changing there.

775

Um, so yeah, there's a lot that

goes into it, but it's super

776

interesting how they built that.

777

So yeah, that's when we look at

like unstructured, like a video,

778

which is, as you said, just

like this long rambling thing.

779

Um, we're going to have to infer some,

put some structure onto it versus

780

just like break up every 200 words.

781

Like

782

that's probably too

783

Randall Stevens-1: Yeah, this

is where, you know, the context

784

becomes everything, right?

785

The better, the better context you can

put around all this information, the more

786

accurate your results are going to be.

787

Evan Troxel: I think, you know,

to the question or to the topic

788

about, I mean, you're using a great

example here with programming, right?

789

I think of architect programming, clients

don't even know what that means, but

790

Christopher Parsons: Right.

791

Evan Troxel: that word in front of them.

792

Um, and, and they, they learn what

it means going through the process.

793

But if you were to bring up programming,

they would never guess that it's about

794

spaces and adjacency and size and scope

and all those things and square footage.

795

Like, they wouldn't, they

wouldn't think about that.

796

And my question is, is, The older

generations in our firm who are the

797

ones with a lot of that encapsulated

798

Christopher Parsons: Hmm.

799

Evan Troxel: like how are

you getting that out of them?

800

Because a lot of this just

happens through conversation.

801

It's not recorded.

802

It's, and so, I mean, do

you have advice for firms on

803

Christopher Parsons: Hmm.

804

Evan Troxel: capture this information

moving forward to get that?

805

put into these models so that these

models are smarter because again,

806

like a lot of this is just spoken.

807

It's never recorded in

or maybe it's in an email

808

Christopher Parsons: Sure.

809

Evan Troxel: and, and, and trying to like

dig all that up and get it in there to

810

train this my firm would be really hard.

811

I mean, are you just doing that on behalf

of everybody with your opt in firms AEC

812

Christopher Parsons: Yeah, it's a great,

I mean, it's a great segue, right?

813

So we are offering two different

embeddings models to our clients.

814

So for the folks who opt out, which

is the default, they'll just use the

815

generic open source embeddings model.

816

And for the people who open, who

opt in and contribute content,

817

they'll use the AEC specific model.

818

Um, but I want to come back, go back

to your question a little bit, because

819

I think it's a super important one.

820

Yeah.

821

It's one of the reasons

we're building an LMS.

822

I mean, we're obviously primary reason

we're building an LMS is to help our

823

clients grow and develop their people.

824

and it turns out that, you know,

what's really important, isn't just

825

volume of content for what we're trying

to do in the kind of search case.

826

I showed, we want high quality content

and we want a lot of it, but we want

827

the quality parts really important.

828

And so.

829

When people generally take the time to put

together, and you know, this could just

830

be a simple hour recorded zoom on Zero

carbon buildings or something like that.

831

Or, like how to work with healthcare

and like, here are the main phases and

832

like, we're gonna talk about programming

is one of them or something like that,

833

what I have observed in my career

and why watching our clients is that,

834

you know, the people that put those

things together, put a lot of thought

835

and care into them, and then there

are interesting discussions that kind

836

of come out in the Q and a, and so.

837

It's why video is such a core technology

for us, because it's easier for them

838

to do that, you know, Zoom meeting

and kind of share their knowledge,

839

then sit down and write a guide or

write a document or write a long post.

840

Um, and then if they can, and this

is why the LMS, now it's like, okay,

841

but then we can reuse that content.

842

It's not just the Lunch and Learn

that happened on a Tuesday in March.

843

This is now a course that can be for

future employees that come along that

844

want to learn about, you know, content.

845

You know, net zero projects, and

it can be assigned, or it can

846

be recommended depending on your

career track or your learning path.

847

And so we get more ROI out of those

videos in terms of educating folks, but

848

then we get more high quality content

we can use to answer people's questions,

849

but then also train these models to

850

understand that AEC

851

Randall Stevens-1: I wish both of you

guys had been able to be in New York

852

with us last week for the Confluence

853

event that we had.

854

But we had, two of our speakers were from

Thornton, the core studio at Thornton

855

Tomasetti, and, um, it actually wasn't

part of their, uh, Uh, presentation, but

856

afterwards, as part of the discussion,

there was a, uh, a gentleman that

857

worked at Thornton Tomasetti for 30

plus years who passed away recently.

858

And, uh, they had built an AI engine.

859

He had, uh, he had documented, like, uh, I

think they threw out the number, like 10,

860

000 interactions that were through either

through email correspondence or Q and A.

861

So he was literally the, you know,

862

the, the, the person that knew the

863

most.

864

you know, it's a really beautiful

thing when you think about it as,

865

as, you know, how do you capture that

knowledge and experience in such a way

866

that can get to the next generation?

867

Cause it's, you know, and I think the,

you know, what you're working on, Chris,

868

the, the, this kind of interaction

with that kind of information is how

869

can learnings pass on to the next

870

gen and we don't lose, we

don't lose what was valuable.

871

And we got into the

discussion about, you know.

872

In the end, when you get a lot

of information, it's like, what

873

might've been true 20 years ago

874

may not be true today.

875

So you have to fight that part of it.

876

The other piece I'll throw out, and I'd

love to engage with your, uh, in this

877

kind of thinking, but I've been throwing

out, it's like a lot of the work that

878

everybody's doing on AI is, you know,

we have a lot of information in this

879

industry, especially in the form of kind

of I'll just call it finished product.

880

It's like, here's what we designed.

881

Here's what this ended up, right?

882

So, so you can easily go and look at that.

883

What's, what seems to be missing

is the, how you got there.

884

It's the, it's the why,

why did you make that

885

decision?

886

Yeah.

887

And so I've been like, um, uh, kind

of throwing out this, like what you

888

really want to start happening is, and

I actually this past weekend post that

889

Confluence event, I actually went on

Amazon looking for, I want a red orb.

890

I want this like device.

891

That when we're back in person with each

other, I wanna like drop that on the

892

table and hit record and it just be this,

it's just gathering the conversation

893

and the why as part of the process.

894

'cause it, I feel like that's part

of, you know, we're gonna, we're gonna

895

get the end result of this stuff,

the polished end result, but we're

896

missing the explanations of that.

897

So anyway.

898

I'd love to hear kinda what your thoughts

about where this is gonna end up.

899

Where

900

does this

901

go?

902

Christopher Parsons: Yeah, I'll

just, I'll give you, so the, in,

903

in, in, I'll give you some KM speak

for what we're describing, which

904

is critical knowledge transfer.

905

And so critical knowledge transfer talks

about, you know, especially when you're

906

talking about, in many cases, it's used

for senior people who were there at

907

key moments, founding, launching new

services or new market types, but it's

908

also when you've got, you know, a super.

909

Rare subject matter expert, and you

want to just be able to like leverage

910

their knowledge to other people.

911

We, we ran an entire day at KA Connect,

uh, our annual conference a few years

912

ago on critical knowledge transfer.

913

We brought in an expert.

914

She was a former professor at MIT

and Harvard and the business school.

915

This was her career.

916

She wrote a book called Deep

Smarts, which I highly recommend

917

her name's Dorothy Leonard.

918

And we ran a project with four of our

clients over the course of six to nine

919

months where Dorothy kind of like used

her critical knowledge method methodology.

920

And we just ran projects in those

companies to kind of pick out a couple

921

of those experts and figure out how

to flesh this deep wise stuff out.

922

Like it's the, it's the deep smarts.

923

It's not like, you know,

simple, simple things like the

924

project details and the square

925

footage and whatever, it's

like, it's wisdom, it's

926

wisdom.

927

There was a story from

Dewberry that was amazing.

928

Very senior, long tenured engineer.

929

And the thing that came out that he had

that just people didn't understand is

930

he had this list of seven priorities

when you're working on a project.

931

And what he said is the important

thing is the order, right?

932

Cause that's where values are determined.

933

And so it's like, you know, for example,

safety, like it doesn't matter, cost,

934

scope, all these different things.

935

It's like, if this thing

is unsafe, We can't do it.

936

And so it's kind of like,

that's a deep way of

937

thinking about how to

938

Randall Stevens-1: Yeah,

It's like a framework,

939

Christopher Parsons: It's a framework,

940

you're right.

941

You want to excavate these frameworks

that a lot of times these experts

942

don't even know that they have, right.

943

They're implicit and then they're

in the back of their brain, but

944

they don't know that they use them.

945

And so that's what this kind of process

of critical knowledge transfer is

946

about, is about excavating those things.

947

And they're oftentimes done

through project stories is how

948

you get them out of people.

949

Right.

950

You say like, okay, well,

you did this on this project.

951

What did you consider?

952

What alternatives did you look at?

953

How did you make the

decision that you did?

954

And through that process of giving

those exact examples, that's where you

955

start finding out those kinds of hidden

frameworks that people are using and their

956

heuristics that they're not

957

even

958

Randall Stevens-1: yeah, I

think about, um, you know,

959

especially I'm a great example.

960

I'm not right out of school,

but I never practiced.

961

So it's like I've got, you

know, went to architecture

962

school, but I never practiced.

963

So I'm, you know, I don't have

all of that deep knowledge.

964

You know, knowledge, practical

knowledge about some of this

965

kind of get it in theory.

966

But, and I, I've used this example.

967

I have a good buddy of mine that

I went to architecture school with

968

who now owns a firm practicing.

969

And I had this, I had this

PDF with details on it.

970

To me, they're just details.

971

As soon as I showed it to him, he

starts pick, you know, picking at them

972

and explaining different parts of it.

973

And I've used that as the example of like,

If I, if, if I was the, I'm proverbial,

974

I'm like perpetually a 24 year old in

my knowledge of architecture because

975

I never did really progress from the, you

know, I understand conceptually what it

976

is, but I don't know, I'm not intimate

with that, but what, uh, we're actually

977

building some tools, uh, we're building

some annotation tools in avail and I'm

978

very, right now it's kind of redlining

and the traditional, but I'm very keen on.

979

Uh, starting to capture video and

audio as part of the annotation

980

process, because what I'd really

981

like to see is,

982

Evan Troxel: It's

983

Randall Stevens-1: want,

when you're marking it up, if

984

you'll just record this and

985

explain it at, in the

moment, why you're doing it.

986

To me, it seems like

987

that's, we've never really had that.

988

So I'm kind of keen on starting

to experiment with will people

989

be willing to just record an audio?

990

And I think Chris, the work you're doing

is, is the right, is, is the right way to

991

think about this, that people, they don't

want, they don't have to go type it up.

992

I'm not going to go write a blog post

about it, but I'll, while I'm doing

993

this, I'll talk and, um, That's the

easiest form of me getting this,

994

uh, information out and then it

can be chewed on and perpetually

995

available, you know, in perpetuity.

996

It's like,

997

Christopher Parsons: no, you, you, you,

so I mentioned there were four firms as

998

part of this critical knowledge project.

999

One of the other ones was Shepley

Bullfinch and Jim Martin and his

00:46:39,551 --> 00:46:41,111

team did a couple experiments.

00:46:41,111 --> 00:46:42,951

One of them was with kind of the senior.

00:46:43,351 --> 00:46:48,741

You know, quality CA kind of person who

is reviewing sets and they recorded him

00:46:48,741 --> 00:46:50,411

with a junior person reviewing sets.

00:46:50,411 --> 00:46:53,361

And as he's redlining, the role

of the junior person is saying

00:46:53,361 --> 00:46:54,661

like, well, what did you see there?

00:46:54,761 --> 00:46:55,781

Why did you circle that?

00:46:55,811 --> 00:46:57,171

You're like, why did you make that note?

00:46:57,431 --> 00:46:59,231

Because the expert

doesn't necessarily know.

00:46:59,231 --> 00:47:00,521

It's just so intuitive.

00:47:00,581 --> 00:47:02,241

Like it's tacit knowledge at this point

00:47:02,416 --> 00:47:02,736

Randall Stevens-1: right.

00:47:03,411 --> 00:47:03,601

Christopher Parsons: and They

00:47:03,626 --> 00:47:04,726

Randall Stevens-1: think

you should know it, right.

00:47:04,726 --> 00:47:05,146

Because they

00:47:05,146 --> 00:47:05,526

forgot.

00:47:06,206 --> 00:47:06,996

Christopher Parsons: I

think you should know it.

00:47:07,036 --> 00:47:08,606

So that was one of the

experiments they did.

00:47:08,606 --> 00:47:11,296

And then they did this other cool one

with one of their healthcare experts

00:47:11,296 --> 00:47:15,446

where they would show a picture of,

uh, uh, like an operating room or

00:47:15,446 --> 00:47:18,556

something like that, and they would

ask her to say like, what's wrong.

00:47:18,921 --> 00:47:20,981

You know, like what's broken

here, you know, and she'd say

00:47:20,981 --> 00:47:23,081

like, well, that thing's not

decode, that thing's not decode.

00:47:23,081 --> 00:47:23,761

I'm looking at that.

00:47:23,761 --> 00:47:24,741

Someone's going to trip on it.

00:47:24,741 --> 00:47:28,571

And like, just like the knowledge of

how she can just process one simple

00:47:28,581 --> 00:47:30,641

photo and just like tear it apart.

00:47:30,661 --> 00:47:32,871

Like that turned out to be super useful.

00:47:32,941 --> 00:47:36,131

And so they like turn that into some,

that's how you start excavating some of

00:47:36,131 --> 00:47:36,471

those things.

00:47:36,481 --> 00:47:36,601

So

00:47:36,601 --> 00:47:36,761

those

00:47:36,906 --> 00:47:37,706

Randall Stevens-1: I think, uh,

00:47:39,453 --> 00:47:42,003

Evan Troxel: exposing that

pattern recognition into

00:47:42,061 --> 00:47:42,431

Christopher Parsons: Yeah.

00:47:42,431 --> 00:47:44,251

Mm hmm.

00:47:44,251 --> 00:47:45,161

Mm

00:47:45,563 --> 00:47:45,913

Evan Troxel: deal.

00:47:46,043 --> 00:47:47,573

And I was just going to

say, Randall, about like the

00:47:47,583 --> 00:47:48,713

markup and talking out loud.

00:47:48,733 --> 00:47:51,663

This is actually something that you

could do in a zoom meeting and you

00:47:51,663 --> 00:47:54,063

don't even have to have somebody

else in the zoom meeting, right?

00:47:54,263 --> 00:47:58,953

You, because you have a whiteboard, you

can load up a PDF and you can record

00:47:58,953 --> 00:48:00,393

it to the cloud at the same time.

00:48:00,443 --> 00:48:03,183

And now with the zoom, I think

they're using Otter's engine to do

00:48:03,541 --> 00:48:03,841

Christopher Parsons: hmm.

00:48:04,573 --> 00:48:05,143

Evan Troxel: and.

00:48:05,458 --> 00:48:11,058

It just seems like a great way to create,

start creating a library of captures that

00:48:11,078 --> 00:48:17,168

are video plus markup, plus transcription,

takeaways, all those things that you could

00:48:17,528 --> 00:48:21,328

potentially use to train your staff in the

00:48:21,328 --> 00:48:22,378

future because

00:48:22,446 --> 00:48:22,656

Randall Stevens-1: Yeah.

00:48:22,656 --> 00:48:26,696

I think the trick to this is all

going to be making it, um, back as,

00:48:26,736 --> 00:48:29,896

as part of the normal workflow, not

00:48:29,896 --> 00:48:32,486

be something that you go do afterwards.

00:48:33,131 --> 00:48:33,531

right?

00:48:33,561 --> 00:48:36,311

but, but it's not, but, but

maybe it's not too difficult.

00:48:36,351 --> 00:48:37,131

Like, hey, just

00:48:37,181 --> 00:48:38,811

verbalize what you're thinking and

00:48:38,811 --> 00:48:39,521

say it out loud,

00:48:39,601 --> 00:48:39,821

right?

00:48:39,981 --> 00:48:41,311

Christopher Parsons:

I think it's, I agree.

00:48:41,311 --> 00:48:44,361

I think it's not hard because I

think, you know, in, in kind of,

00:48:44,401 --> 00:48:47,971

you know, mindset is the most

important thing for behavior change.

00:48:47,971 --> 00:48:50,541

And I do believe in

general in this industry.

00:48:50,951 --> 00:48:54,791

We have a culture of mentorship and

wanting to advance the next generation.

00:48:54,811 --> 00:48:58,671

And so I think this is just a different

technique, but it fits into a through

00:48:58,671 --> 00:49:00,011

line of something people already

00:49:00,011 --> 00:49:00,561

want to do,

00:49:00,751 --> 00:49:01,111

which is

00:49:01,191 --> 00:49:01,851

Randall Stevens-1: altruistic.

00:49:01,851 --> 00:49:02,331

Yeah, it's a

00:49:02,331 --> 00:49:02,651

very

00:49:02,871 --> 00:49:02,991

you

00:49:02,991 --> 00:49:03,131

know,

00:49:03,431 --> 00:49:04,261

Christopher Parsons: it's very altruistic.

00:49:04,261 --> 00:49:05,081

It's a great profession.

00:49:05,228 --> 00:49:06,228

let's talk about altruism.

00:49:06,228 --> 00:49:08,218

That's a great segue to community AI.

00:49:08,328 --> 00:49:11,468

Um, so let me tell you a

little bit about what we did.

00:49:11,478 --> 00:49:16,908

So when we saw this opportunity to

build these AEC specific models, um,

00:49:17,668 --> 00:49:18,548

we're like, how are we going to do this?

00:49:18,548 --> 00:49:19,588

It has to be opt in.

00:49:19,903 --> 00:49:21,923

You know, we're not just going to

start scraping everyone's data.

00:49:22,223 --> 00:49:25,463

Um, but we saw that there was this

great kind of quid pro quo, right?

00:49:25,463 --> 00:49:28,733

If you help share some data to help

us train these models for other folks

00:49:28,753 --> 00:49:32,063

in our community, you'll be able

to benefit them, benefit from them.

00:49:32,413 --> 00:49:35,843

So the two models we're building are the

embeddings model, which we just talked

00:49:35,853 --> 00:49:39,812

about a little bit for more relevant

search results, and the transcription

00:49:39,813 --> 00:49:41,753

model for more accurate video captions.

00:49:42,463 --> 00:49:44,843

And so our clients are

opted out by default.

00:49:45,383 --> 00:49:47,463

Once they join it, they

get access to the models.

00:49:48,213 --> 00:49:53,683

So at a high level, we're extracting

what we're after, our AEC specific

00:49:53,683 --> 00:49:58,273

terms, phrases, and then the context

for those terms and phrases from

00:49:58,273 --> 00:50:00,023

content at the firms that participate.

00:50:00,513 --> 00:50:05,903

So, you know, here's a word soup example,

you know, anything from acronyms to

00:50:05,973 --> 00:50:08,703

products, to different kinds of systems.

00:50:08,703 --> 00:50:12,403

Like these are the things that,

that generally the embeddings model

00:50:12,403 --> 00:50:15,323

out of the box, the generic one

and a generic transcription, video

00:50:15,323 --> 00:50:16,803

transcription model don't do that well.

00:50:17,298 --> 00:50:20,588

So that's what we use to fine

tune both the embeddings model

00:50:20,638 --> 00:50:21,748

and the transcription model.

00:50:22,808 --> 00:50:26,738

At a high level, what we're really trying

to do is kind of build an AEC crowdsource

00:50:26,748 --> 00:50:28,748

dictionary from our clients, right?

00:50:28,748 --> 00:50:30,648

So that we can do more smart things with

00:50:30,648 --> 00:50:30,988

AI.

00:50:31,268 --> 00:50:31,658

Randall Stevens-1: makes sense.

00:50:32,651 --> 00:50:32,931

Christopher Parsons: Yep.

00:50:33,083 --> 00:50:35,053

Evan Troxel: even people are going

to use different words for the same

00:50:35,211 --> 00:50:35,661

Christopher Parsons: Yes.

00:50:36,551 --> 00:50:36,821

Yeah.

00:50:36,973 --> 00:50:39,653

Evan Troxel: Different people

are going to call gypsum

00:50:39,891 --> 00:50:40,291

Christopher Parsons: Right.

00:50:40,713 --> 00:50:43,903

Evan Troxel: drywall, or gypsum wall

board, or, there's going to be an

00:50:43,903 --> 00:50:47,463

abbreviation for that, which ties

back into content management, like

00:50:47,463 --> 00:50:48,693

what Randall's tackling, right?

00:50:48,693 --> 00:50:51,893

It's like, you would abbreviate it on a

detail differently than you would even say

00:50:51,941 --> 00:50:52,441

Christopher Parsons: Right.

00:50:52,441 --> 00:50:53,031

Right.

00:50:53,183 --> 00:50:54,033

Evan Troxel: of these things have to

00:50:54,101 --> 00:50:54,941

Christopher Parsons: They

have to tie together.

00:50:54,941 --> 00:50:58,031

And that's what's the beauty of

a vector search versus keyword

00:50:58,041 --> 00:50:59,421

back to the original piece, right?

00:50:59,421 --> 00:51:04,111

Because vector, it's about relative

proximity, not exact match, right?

00:51:04,121 --> 00:51:07,081

So these terms are in

a neighborhood, right?

00:51:07,381 --> 00:51:09,931

These terms are all kind of

in a neighborhood together.

00:51:10,821 --> 00:51:13,151

And like these terms are all kind

of in a neighborhood together.

00:51:13,151 --> 00:51:15,531

They're not exactly

synonymous, but they're close.

00:51:15,531 --> 00:51:18,521

And like, that's how we start

understanding the relationship between

00:51:18,631 --> 00:51:19,981

different ways of saying the same thing.

00:51:20,284 --> 00:51:22,004

Um, the last thing I want

to say on the search front.

00:51:22,329 --> 00:51:26,189

And if we still have time, it'd be

good to go into, uh, training the

00:51:26,199 --> 00:51:30,539

video transcription model a little bit

is, um, our vision for the product.

00:51:31,149 --> 00:51:34,529

Is to be able to answer questions

using your highest quality source

00:51:34,529 --> 00:51:37,609

of firm wide knowledge, whether

that lives in Synthesis or not.

00:51:37,959 --> 00:51:39,579

And so we do have a history of doing this.

00:51:39,579 --> 00:51:42,289

There's obviously Synthesis

content and the Synthesis LMS

00:51:42,299 --> 00:51:43,909

content when it's released.

00:51:44,199 --> 00:51:48,119

We've got integrations with ERP and

CRM systems from our directories

00:51:48,119 --> 00:51:49,379

and profiles that we saw.

00:51:50,059 --> 00:51:53,269

We integrate with OpenAsset for

imagery and Newforma for contacts.

00:51:53,299 --> 00:51:54,719

And we do Zendesk for Help Center.

00:51:54,719 --> 00:51:56,979

So that's, this is all stuff we do today.

00:51:57,359 --> 00:52:01,009

So if you were to search Synthesis

Today on work sharing, you can see

00:52:01,009 --> 00:52:04,419

that one of the sources we pulled

in was from a Zendesk Help Center

00:52:04,429 --> 00:52:06,029

article on Revit work sharing.

00:52:06,779 --> 00:52:13,549

So we just have pretty ambitious plans

to expand and make more, um, more sources

00:52:13,549 --> 00:52:15,169

of that firm wide knowledge available.

00:52:15,559 --> 00:52:17,349

So that might be something like Teams..

00:52:17,654 --> 00:52:20,534

Like looking at kind of

messages and documents from

00:52:20,534 --> 00:52:22,374

public channels within teams.

00:52:22,804 --> 00:52:25,774

That might be something like Freshworks,

which is a Zendesk competitor.

00:52:25,774 --> 00:52:28,294

And it's like outside of Zendesk,

it's the most common help center

00:52:28,294 --> 00:52:29,864

software we're seeing in our community.

00:52:30,634 --> 00:52:35,194

Um, we'll likely going to build a search

connector API that will allow clients

00:52:35,194 --> 00:52:38,474

to build their own integrations, to

push content into Synthesis Search.

00:52:38,964 --> 00:52:41,814

And we're looking at this idea

of a public website indexer.

00:52:41,824 --> 00:52:44,334

So we've kind of gotten

two use cases for this.

00:52:44,344 --> 00:52:46,414

One is to index their own website.

00:52:46,974 --> 00:52:50,574

It's very common that clients will do

high quality content on their website

00:52:50,574 --> 00:52:52,064

that doesn't make its way into Synthesis.

00:52:52,679 --> 00:52:55,689

And then in the second use case,

it's to index the help centers

00:52:55,689 --> 00:52:56,979

of trusted software partners.

00:52:56,979 --> 00:53:00,789

So maybe a deck, uh, you know, indexing

the avail help center or building

00:53:00,789 --> 00:53:04,069

some kind of integration with, uh,

the avail help center, the open asset

00:53:04,069 --> 00:53:06,509

help center, or the Autodesk help

center, or whatever it might be.

00:53:06,589 --> 00:53:08,579

So these are things we're exploring.

00:53:08,579 --> 00:53:11,529

We're definitely committed to building

the search piece I showed you.

00:53:11,529 --> 00:53:13,419

And then the teams and the

fresh work and the search

00:53:13,419 --> 00:53:14,879

connector API and public website.

00:53:14,879 --> 00:53:18,429

These are all future explorations

that we're trying to validate and

00:53:18,429 --> 00:53:19,509

see if this is worth building.

00:53:19,619 --> 00:53:20,149

so the.

00:53:20,560 --> 00:53:24,280

um, the idea for that is this is

00:53:24,430 --> 00:53:25,100

So by the end of the

00:53:25,100 --> 00:53:25,290

year.

00:53:25,480 --> 00:53:26,050

Randall Stevens-1: That's great.

00:53:26,450 --> 00:53:30,260

Are you thinking, Chris, because we're

kind of, because we're experimenting

00:53:30,260 --> 00:53:33,510

with a bunch of these different kind

of search methodologies and changes in

00:53:33,800 --> 00:53:38,150

search methodology, do you envision,

like, with Synthesis, are you all working

00:53:38,150 --> 00:53:43,680

through that this can be a single search

box, like the Google search box, that

00:53:43,680 --> 00:53:45,390

you'll begin everything from that?

00:53:45,390 --> 00:53:51,830

Or do you have to, like, quickly,

quickly put people in proper

00:53:51,830 --> 00:53:52,840

directions to get to that?

00:53:53,180 --> 00:53:56,010

To the right info because if you do

connect all of those things that you're

00:53:56,010 --> 00:53:57,670

talking about it can get noisy again,

00:53:58,100 --> 00:53:58,470

right?

00:53:58,470 --> 00:54:01,690

Yeah.

00:54:01,710 --> 00:54:04,340

Christopher Parsons: a very good question

and I don't know all the answers yet.

00:54:04,340 --> 00:54:06,010

I'll just tell you a couple of