Latest / The Jon and Marc Podcast / Ironman, Not Terminator: The Reality of AI in Modern Plaintiff Law

Transcript

Ironman, Not Terminator: The Reality of AI in Modern Plaintiff Law · The Jon and Marc Podcast

• 0:00
  We would say it's more like an Iron Man for this function than
• 0:03
  like a Terminator, because if you want the data that goes into
• 0:06
  the downstream system to be perfect, some elements of that
• 0:08
  will require human review. Every decision our software
• 0:11
  makes, whether it's what matter it goes into what type of
• 0:14
  document it is, has to have a confidence score.
• 0:16
  So if it's not confident and a real confidence score grounded
• 0:19
  in reality, it's directing a person to review that to make
• 0:22
  sure. Welcome to the John and Mark
• 0:33
  podcast. We are live in Miami with Vivic,
• 0:37
  the CEO and founder of Foundation AI, and we're very
• 0:40
  excited to have him on the podcast here.
• 0:43
  Fun fact, we met with your team before this podcast and did some
• 0:46
  research. You got some background about
• 0:49
  you, what you like, what you don't like and some tips on how
• 0:51
  to properly interview. So I still like my team after I
• 0:54
  hear what they told you. They gave us some dirt.
• 0:56
  We asked about you and they were ears Perk and they all chimed
• 1:00
  in. All joking aside, we're very
• 1:02
  excited to have you on. Unfortunately, we had to
• 1:04
  reschedule this original podcast and we're so happy we're able to
• 1:08
  get you on finally. We've been looking forward to
• 1:10
  this. Thank you for joining us in the
• 1:12
  John Mark. Podcast, thanks so much for
• 1:14
  having me, guys. I'm really excited to be here.
• 1:15
  Before the podcast started, we were chatting a little bit and
• 1:19
  you talked about something that I really didn't think you guys
• 1:21
  did. That's a bigger issue than I
• 1:23
  probably realized. E-mail so people think of
• 1:26
  foundation, John and I thought like dealing with.
• 1:29
  Physical mail, Yeah, yeah, I called it like the AI mail room.
• 1:32
  But one thing that you guys are doing helping a ton on is
• 1:35
  dealing with emails. Yeah.
• 1:36
  Can you talk to me about that? Because once again, that is the
• 1:39
  bane of my existence. Yeah, yeah.
• 1:40
  We, we have a rule. By the way, at our company,
• 1:43
  there is no such thing as an e-mail between employees.
• 1:46
  Yeah, it does not exist. It happens in Slack.
• 1:48
  If someone starts a conversation in e-mail, we we ping them on
• 1:52
  Slack. We're like, Nope, we don't do
• 1:54
  that. So imagine amidst all those
• 1:56
  emails that everyone's getting at a firm where they have
• 1:58
  hundreds of people that buried in those are emails coming in
• 2:01
  from third parties that have attachments.
• 2:03
  So like medical records, bills, all this stuff is being emailed
• 2:06
  not just to, you know, the handling attorney, but to the
• 2:10
  case managers, legal assistance, like all over the place.
• 2:13
  And so historic when we started the business, which was like 8
• 2:15
  years ago now, like a lot of it was physical mail firms would
• 2:18
  get pre COVID, most of the documents coming into a firm
• 2:21
  with physical mail or like fax, which also would go to an e-mail
• 2:24
  box. Nowadays, more and more
• 2:26
  documents are coming in through all these different channels.
• 2:29
  So every one of those employees at one of the firms, some firms
• 2:32
  have hundreds of employees, every one of them is receiving
• 2:34
  documents, right? And any one of those documents,
• 2:37
  if it doesn't make it into the downstream practice management
• 2:39
  system, that's a problem. And so one of the things we do
• 2:43
  is we integrate with all those e-mail boxes, not just the
• 2:45
  centralized ones built to like receive faxes or court notices,
• 2:49
  but also we can integrate with all the individual e-mail boxes
• 2:52
  across all the attorneys staff, whoever the firm wants, and any
• 2:56
  third party attachment that comes in.
• 2:57
  So any e-mail that comes in with an attachment automatically pull
• 3:00
  it into the system, process it, put it into the practice
• 3:03
  management system through the same process as all the physical
• 3:05
  mail. So any document, regardless of
• 3:07
  how it comes in, essentially our job is to make sure that we
• 3:10
  capture all documents and they all get treated in the same,
• 3:13
  like standardized workflow name the same way, save the right
• 3:16
  place, the same people get alerted so that no matter how
• 3:19
  they come in the firm, you know, process them the same for the
• 3:22
  firms that go with us, that whole standardization, that
• 3:24
  certainty is, is what they're buying, right?
• 3:26
  Is it just the attachments or do you guys actually crawl the
• 3:29
  emails? Yeah.
• 3:30
  So right now we're just focused on the attachments because our
• 3:32
  main job is to capture all the documents that come into a firm
• 3:36
  and make sure they're in like the right place.
• 3:39
  In the future, we might start thinking about like the e-mail
• 3:41
  boxes, helping to process those more.
• 3:42
  But right now, like our job is really to get attachments or
• 3:46
  physical like documents, right? Medical records, bills, court
• 3:51
  notices, correspondence from the carriers, all the stuff that
• 3:54
  comes in. Our job is to get it into the
• 3:56
  right place in the client's system, regardless of the
• 3:59
  channel it entered the firm in, named their naming conventions
• 4:02
  saved in the right folder and execute the workflows that they
• 4:05
  determine. So firms will have, you know,
• 4:07
  all sorts of different alerting mechanisms that they'll, they'll
• 4:10
  set up or try to set up before we get there.
• 4:12
  And once they have like Foundation AI, you know, for
• 4:15
  each document type, they can figure out exactly what naming
• 4:18
  convention it should have, what folder it should go into in
• 4:20
  their subfolders and their practice management and who on
• 4:23
  the team, you know, would get should get tasked alerted how
• 4:26
  that task should be executed and all that.
• 4:29
  And we automate that kind of work.
• 4:30
  Do you guys see a lot more so e-mail?
• 4:32
  I didn't think you guys did but that makes perfect sense to me
• 4:35
  as a huge Ave. where all the documents are coming in for its
• 4:38
  physical mail nowadays. Do you guys see like a new
• 4:40
  influx of people sending and stuff through like text messages
• 4:43
  or different like mediums outside of just e-mail now?
• 4:47
  I think that firms are getting more and more of this stuff
• 4:50
  through both their portals, through text and all that.
• 4:53
  Generally what we're dealing with is not so much the
• 4:55
  correspondence between like the the actual client and the firm.
• 5:00
  Usually it's after their matter is set up and they have things
• 5:02
  then, then we're really dealing with all the correspondence
• 5:05
  coming in, correspondence documents from all the third
• 5:08
  parties once the matter is set up.
• 5:09
  But to your question, I think that there's way more stuff
• 5:12
  coming in to firms and we see it in their files through text
• 5:16
  messaging and things like that. Certainly pictures and stuff
• 5:19
  like that are coming in that way.
• 5:20
  Do people still fax? Yeah, that's still a thing.
• 5:23
  That that you do know that the only way you can send documents
• 5:26
  to the IRS is by faxing it to them.
• 5:28
  I I did not know that now you. Know it's, it's insane.
• 5:31
  So the the US healthcare system, how much healthcare data, you
• 5:36
  know, the, the promise of interoperability of healthcare
• 5:39
  systems from like 2009. Like still to this day, most
• 5:42
  healthcare data is transmitted via fax.
• 5:45
  Like it's this misnomer that fax is the only thing that's HIPAA
• 5:48
  compliant, even though now like emails are encrypted enough to
• 5:51
  be as well. But tons of documents, usually
• 5:55
  it's healthcare documents are still going via fax.
• 5:57
  But in a lot of firms like like in California work comp, for
• 6:00
  example, law firms are still faxing things to each other.
• 6:03
  And and it's. Yeah.
• 6:05
  If I if I showed my kids a fax machine, they would have
• 6:07
  absolutely no idea what the hell that thing.
• 6:09
  Is so much of it gets transmitted into like an E fax
• 6:12
  and goes to an e-mail box, but we still, but we've still seen
• 6:15
  clients that like full on print out a stack like the faxes come
• 6:19
  out and then they scanned it in. It's like the control.
• 6:23
  Let's just kill trees. Yeah, that's what they're all
• 6:25
  about. It's out of control.
• 6:26
  It's super fascinating. I want to get take a little step
• 6:29
  back what you do and kind of get your view on the industry
• 6:33
  because you know, we're John and I were talking about we walk
• 6:36
  around these conferences every year.
• 6:37
  We go to all of them and this NTL, all the booths.
• 6:41
  They used to be medical Med records companies, intake
• 6:44
  companies, marketing companies, some other service companies.
• 6:48
  And now it's all soft like CRMS and AI platforms.
• 6:52
  Yeah, it's all software. Yeah.
• 6:54
  AI is like the big boom starting last year.
• 6:56
  Now it's like you guys were everywhere.
• 6:58
  Yeah. What's your take on the industry
• 7:01
  and, and to the legal space and like how do you see foundation
• 7:05
  playing in that or or dealing with the noise of all the
• 7:08
  different AI stuff that's going? On Yeah, it's, it's a great
• 7:10
  question and I I talk about this a lot when I speak at these
• 7:13
  things. You know, I think that, well,
• 7:15
  first, we've been doing this for for a while, right before AI was
• 7:19
  AI before ChatGPT came out from the problem that I saw the firm
• 7:23
  I worked at, but OCR has been around for a while.
• 7:25
  So I can take you back in the history of like how this all
• 7:27
  came about. But what really happened is, is
• 7:29
  COVID, I mean, in 2020 COVID hit and basically the legal industry
• 7:35
  moved like light years in two years, but not all the legal
• 7:38
  industry. Like if you think about it, most
• 7:41
  of this AI is all about making people more efficient right now,
• 7:44
  who wants to be more efficient? Not the defense side, not like
• 7:48
  the the side that bills ours, but the side that does
• 7:51
  contingency work. Consumer lawyers, plaintiff
• 7:53
  lawyers, anyone doing flat fees contingency.
• 7:55
  Like you're incentivized to be more efficient.
• 7:58
  The other thing was a lot of the firms I think like weren't using
• 8:02
  like only the large ones I think pre pre 19 or 18 were were on
• 8:07
  like digital systems already. This is when we were at Litify,
• 8:12
  like we were there when Covic hit.
• 8:14
  And that was, I mean, those years for Litify, for file line,
• 8:17
  for, I mean, that's when practice management took off.
• 8:20
  And the reason was a lot of those firms weren't on anything
• 8:23
  yet. So it wasn't like they had to
• 8:25
  RIP and replace like a system they were on.
• 8:26
  They had to, they were on nothing.
• 8:28
  Whereas like a bunch of the other side, they were on these
• 8:31
  weird antiquated systems that were on Prem and they had this
• 8:34
  idea that you store all their data, you know, like in their
• 8:37
  environment and like they had server rooms in their offices.
• 8:40
  But the plaintiff side, a lot of them, they had paper and they
• 8:42
  were like, you know what? Like so they were able to
• 8:44
  leapfrog in a lot of ways and buy cloud based technology
• 8:47
  early, which is when Cleo took off Lidify, Filevine.
• 8:50
  So then you have like the plaintiff side or the consumer
• 8:53
  side. That's always been sort of the
• 8:55
  first mover in tech in technological advancement,
• 8:58
  because they're incentivized to be more efficient, to do things
• 9:01
  better and not to just sit around like billing hours, like
• 9:03
  a lot of the legal kind of market, the broader legal
• 9:06
  industry is. So that starts happening, right?
• 9:09
  And then technology advancements, like naturally, if
• 9:12
  you're trying to be more efficient and you're
• 9:14
  incentivizing more efficient, you'll buy it.
• 9:16
  All right? And so AI now to me and to a lot
• 9:20
  of people, I think if you're looking at it, it's just new
• 9:22
  technology that can allow you to do things that you couldn't do
• 9:25
  with technology before, like even our business, right?
• 9:28
  Like before you could combine OCR has been around, but before
• 9:32
  you could put natural language processing on top of OCR, all
• 9:36
  OCR could do was like standardized forms.
• 9:38
  And even there, it didn't do a great job it.
• 9:39
  Had to be like a standardized Ford.
• 9:41
  Exactly. And if it was any way off, it
• 9:43
  was. Exactly.
• 9:44
  It was called zonal extraction. So it have zones that would be
• 9:47
  highlight and that's like if it's right there, you got it.
• 9:49
  Even if it was tilted it wouldn't work.
• 9:51
  But then when you put natural language processing technology
• 9:53
  on top that you could all of a sudden start reading documents
• 9:57
  like a person and using NLP to do it.
• 9:59
  But you couldn't really do that until you had cloud.
• 10:02
  So in like all this stuff kind of coming together in 2020 when
• 10:05
  you had people that were ready to use cloud based technology
• 10:08
  where you could deploy like AI models because you couldn't run
• 10:10
  them on people's little servers in their in their office.
• 10:13
  So you have that coming in with people that bought newer systems
• 10:17
  that were accessible by API and all that and you had this crazy
• 10:21
  kind of boom in technology. So now if you look at the
• 10:24
  definition of AI, it's just computer systems that do work
• 10:27
  that traditionally were done by humans.
• 10:29
  So at some point, you know, math was done by humans and a
• 10:32
  calculator was AI. Or your spam filter is a great
• 10:34
  example. We used to read every single
• 10:36
  e-mail to try to figure out like what Nigerian Prince was full of
• 10:39
  shit and like what was like. And so that was a human task.
• 10:42
  And now spam filters, you know, you don't see half the crap that
• 10:46
  comes in your box, but we don't think about that as AI.
• 10:49
  It's just how how things are done now, right?
• 10:51
  Similarly, a lot of this stuff, I think as we transition, AI is
• 10:55
  just like tech that can, you know, it's a great word now, but
• 10:58
  really it's just tech that can do stuff that you couldn't do
• 11:01
  before. And so we get to where we are
• 11:03
  now, now with the L, these big LMS and with the ability to like
• 11:08
  approximate or seemingly approximate reasoning.
• 11:10
  You can use this stuff to do all kinds of really exciting things
• 11:15
  like you can draft demands and, and all these documents.
• 11:18
  You can do all the things that some of these big AI players are
• 11:20
  doing. But I think to do that, and this
• 11:22
  gets back to how we play in the space, all of that is only as
• 11:26
  good as the context it has. So if there's no documents or
• 11:30
  there's no information or there's no context, all it's
• 11:33
  going to spit out is garbage. Like if you ask GPT right now
• 11:36
  and you don't, if you ask it a question, you know, I, I use it
• 11:40
  to help with something there today and you don't give a call,
• 11:42
  it'll write a full answer that's really confident based on
• 11:46
  absolutely nothing. You haven't told it anything.
• 11:48
  It'll just spit out a bunch of people like that's not what I
• 11:49
  wanted. So you have to give it something
• 11:52
  for it to process and give you back what you need.
• 11:54
  In the same thing, If there are documents in the data and the
• 11:57
  evidence isn't in the system accessible to the AI, then that
• 12:01
  AI can't do much. So I think where we put it in is
• 12:04
  we're still just this bridge to capture everything that comes in
• 12:09
  and get it to the right place so work can get done, Whether that
• 12:12
  work is done by people and that the job is to put it in the
• 12:15
  folder and assign it to like a paralegal to do something, or an
• 12:18
  AI, whether the job is it goes into this and now an AI is going
• 12:22
  to do the next task. So it's just kind of workflow
• 12:25
  orchestration using technology. Is Foundation built on open AI
• 12:30
  on ChatGPT? Is it built on Gemini?
• 12:32
  Is it built on Claude? It's built on a lot of things.
• 12:35
  You know, I'll give you the lawyer answer.
• 12:36
  It depends. But we actually, we actually
• 12:39
  built a lot of our own models like way before, you know, you
• 12:42
  could use these LMS to do any of this.
• 12:44
  Now we have what we call an ensemble approach.
• 12:46
  So we have multiple models that are working at the same time.
• 12:50
  And one of the reasons that you with what we do, you can't
• 12:54
  really use just an LLM like an off the shelf LLM is goes back
• 12:58
  to what we were just saying LLMS like they still haven't really
• 13:01
  solved the problem of them being be exceedingly confident when
• 13:05
  they have no basis of being confident.
• 13:07
  So in LOM if you ask it how confident it is in something.
• 13:10
  Just read a paper on this. It'll tell you it's confident.
• 13:13
  It'll tell you it's confident even if it's not right.
• 13:15
  And what we do, the way our software works is every decision
• 13:18
  our software makes, whether it's what matter it goes into, what
• 13:21
  type of document it is, what the date of service is on the
• 13:24
  medical record. Every decision has to have a
• 13:27
  confidence score associated because if it's not confident
• 13:31
  enough and a real confidence score grounded in reality, it's
• 13:34
  directing a person to review that to make sure.
• 13:36
  So there are like our system is not a wholly automated system
• 13:40
  that doesn't have people in it. We would say it's more like an
• 13:42
  Iron Man for this function than like a Terminator.
• 13:46
  Because if you want the data that goes into the downstream
• 13:49
  system to be perfect, some elements of that will require
• 13:51
  human review. So with our models, using a
• 13:55
  combination of sort of models we trained ourselves and some of
• 13:58
  the of the the third party LMS, we figured out how to create
• 14:02
  like confidence scores associated with every decision.
• 14:05
  So when and firms can set their own confidence thresholds like
• 14:09
  at the field level at the for these various decisions.
• 14:11
  So for example, if it's a a correspondence that comes in,
• 14:15
  maybe you don't care as much about who the sender is because
• 14:18
  someone's going to read it anyway.
• 14:19
  And so you can accept a lower confidence threshold and have
• 14:23
  more automation and a gig go through without people seeing
• 14:25
  it. But it's a medical record.
• 14:27
  You got to make sure that all those different details that you
• 14:29
  want out of the document are correct.
• 14:31
  So that might show the person at the usually it's the person at
• 14:34
  the firm. Users at the firm use our
• 14:36
  software to kind of optimize this this workflow.
• 14:41
  So it'll show them the stuff that needs their review.
• 14:44
  So if it if there is multiple matters that it could match to,
• 14:46
  there's not enough data on the document to match to one, it
• 14:49
  might show these are the potential matters.
• 14:51
  Or, you know, show them and highlight the field they need to
• 14:54
  examine. Outside of this area, yeah.
• 14:57
  Where do you think the best opportunities are for plaintiff
• 15:02
  law firms today? Start leveraging AI in their
• 15:05
  firm. It's a tough question to answer.
• 15:06
  And because the way I answer it is rather than going out and
• 15:09
  buying AI, what firms should do is they should look at their own
• 15:12
  processes and figure out where there are bottlenecks, like
• 15:16
  where there's where's this process that don't make sense
• 15:19
  and see if there's a solution to, to solve that process.
• 15:21
  If you just go buying like AI, when firms look at their, their
• 15:25
  organization, they might find different firms might have
• 15:27
  different answers. I think that, you know, you
• 15:30
  should look for where you can really optimize your firm and
• 15:34
  buy AI to solve that. I personally think that our
• 15:36
  solution, obviously it's biased, is a really good starting point
• 15:41
  because as they say, like, you know, garbage in, garbage out.
• 15:45
  If you don't have like solid data going into system, then all
• 15:49
  the other AI you want to buy isn't going to be as effective.
• 15:52
  And then also, you know, our software is taking a function
• 15:56
  that everyone can agree like it's 2026, like people shouldn't
• 16:00
  be doing this anymore. It's the kind of thing that like
• 16:02
  people shouldn't be doing. So my attitude on AI is, is, you
• 16:06
  know, I'm sure it will get to the point that it'll be able to
• 16:08
  do even more than it's doing now.
• 16:10
  But but I try to think that people, the firm should look at
• 16:14
  what are things that, you know, people aren't good at doing,
• 16:17
  people don't want to do. And the end consumer, like
• 16:21
  doesn't care if it was done by a person or a machine.
• 16:24
  And the end consumer in some case might be the internal law
• 16:26
  firm user. And like does the attorney when
• 16:29
  they get an alert that this document is there in their
• 16:31
  system and and it's named the right way, do they care that
• 16:33
  like a person did it or not? No.
• 16:35
  And there are lots of things that I think are still uniquely
• 16:38
  human in my mind. Like I personally think that
• 16:40
  firm should try to have people answering phones and talking to
• 16:45
  doing intake. I'm in agreement.
• 16:46
  With that one. Like, I mean, I'm the kind of
• 16:48
  guy, right? I'm pretty tech forward but like
• 16:50
  if I call a number and I get like a machine, I'm the first
• 16:52
  guy just hitting zero like. Give me a completely agree.
• 16:56
  I I'm I'm a loose and if I can't, if I say talk to
• 16:59
  representative and I can't transfer, I'm sick screaming at
• 17:02
  the. Phone and that's what I'm trying
• 17:03
  to change like a flight like imagine like you're trying to
• 17:05
  talk about like a very traumatic experience and try to get and
• 17:09
  and so those are areas where I think humans are uniquely suited
• 17:13
  right and should be doing. And if that person is sitting
• 17:16
  around, you know, filing documents instead of like, doing
• 17:20
  that type of job for your firm, then you know, that person is
• 17:24
  probably not in their highest and best use.
• 17:26
  And so using technology to not necessarily replace people, but
• 17:29
  replace like shitty tasks so that those people can do better
• 17:33
  things and drive more value. What is your response to law
• 17:39
  firm owners who hear AI and think like, I don't want my data
• 17:43
  going out into the cloud and being accessible?
• 17:46
  Like, yes, people say it's secure, but you know, whether
• 17:49
  it's just the LMS training on the data or like the potential
• 17:53
  for data breaches, what can give them comfort that this is a safe
• 17:57
  space to play in right now? If people are thinking this way,
• 17:59
  it's good they're thinking about it correctly because I think
• 18:02
  some firms are just like way too loose with it.
• 18:05
  Like these companies that spun up like a week at half ago and
• 18:09
  they're like, yeah, you have all my data like and they don't
• 18:11
  think about this stuff. Firms should be thinking about
• 18:13
  this. I think, though, as a whole,
• 18:15
  like there are ways to evaluate your vendors to figure out
• 18:18
  whether they're legit or not, right?
• 18:20
  I think a sock 2 type 2 is, is the base is like kind of
• 18:24
  baseline, right? I think about that like, you
• 18:26
  know, I think that firms should vet their vendors the same way
• 18:31
  like people vet their law firms, right?
• 18:33
  So let's say the sock 2 type 2, that's like having a law
• 18:35
  license. It's like that's table stakes,
• 18:37
  right? And then you kind of go on
• 18:39
  beyond that. So do they have references?
• 18:42
  Do they have other customers they work with?
• 18:44
  And that gives you some comfort that that like they're a
• 18:47
  legitimate, you know, enterprise.
• 18:49
  Obviously it's still impossible to like obviate all risk, but I
• 18:54
  don't think anyone at this stage now of the game, most firms that
• 18:58
  are getting breached are getting breached, that have had large
• 19:01
  breaches and a lot of people don't really know about.
• 19:04
  It's less likely that it's like their cloud provider than it is
• 19:08
  that it's something like their own on Prem systems or those
• 19:10
  sorts. Of social engineering or.
• 19:12
  Exactly but but. Even putting social.
• 19:14
  Engineers are crazy. Yeah, forget about like the the
• 19:17
  breaching side of it. Like what about a firm that just
• 19:19
  doesn't want their? What doesn't want models being
• 19:22
  trained on their own data? I think it depends on what type
• 19:24
  of model and firm should should think about that, right?
• 19:26
  So for example, like we don't like we get everyone's data that
• 19:30
  goes to our system, right, But we're not building models on the
• 19:33
  outcomes or models on sort of the contents of those documents.
• 19:37
  Now our models might get better at reading documents.
• 19:40
  So based on, you know, your document, it might get better at
• 19:43
  figuring out where the, you know, data services on a
• 19:46
  document the same way. But to that point, if you're not
• 19:50
  willing to let the model improve on training, then you can't get
• 19:56
  a model that's already improved and the model is not going to
• 19:58
  get any better. So I think that most of most
• 20:01
  providers, you know, are using aggregated and anonymous data to
• 20:05
  improve models. If you're clear about what
• 20:07
  models you're improving with, with the data and it's not
• 20:11
  taking like, you know, the, the contents of the documents as
• 20:15
  opposed to using the documents, the the form of the documents
• 20:19
  versus substance of the documents.
• 20:20
  I think there's a discussion there.
• 20:22
  You know, I would also ask like. Why?
• 20:25
  What makes people uncomfortable about it?
• 20:27
  It's like if a lawyer came and you know, worked for you and
• 20:30
  went somewhere else, they're going to have learned as a by
• 20:33
  virtue of having worked for. You it's less about like the
• 20:35
  processes and more and more about like you're saying the
• 20:37
  substance, right? Like if there's medical records
• 20:39
  that are getting attached to an e-mail or coming in through
• 20:41
  physical mail and those are getting scanned and those are
• 20:43
  getting run through a model, like now you've got potential
• 20:46
  HIPAA violations, right? If that stuff winds up getting
• 20:49
  exposed or or gets surfaced in someone else's response or the
• 20:52
  model starts hallucinating, you know.
• 20:54
  I will say something, because this is a common misnomer for
• 20:57
  plaintiff attorneys. Technically, HIPAA doesn't
• 20:59
  apply. Like unless you're representing
• 21:01
  healthcare providers and covered entities, you're not ABA.
• 21:03
  You're not ABA. Yeah, yeah, but but, but still
• 21:06
  though, you should make sure that any vendor you trust is
• 21:09
  HIPAA compliant because it's kind of the goal set for
• 21:11
  medical. Yeah, I'm sure there's like
• 21:12
  Social Security numbers and some of those documents are like,
• 21:15
  yeah, what was the term that you used for like the model multiple
• 21:18
  models, ensemble the ensemble models, right?
• 21:21
  I'm assuming some are custom, some are are off the shelf,
• 21:24
  we'll call them, right? Like standardized models, when
• 21:27
  those models that are not custom wind up getting updated.
• 21:30
  So when ChatGPT went from three to four goes from 4:00 to 5:00
• 21:33
  or five to five two, right, They improve, they're getting trained
• 21:37
  on their own data sets, right? Do you go through a process of
• 21:41
  updating like your system sort of in lock step to as these
• 21:45
  models update consumers, the law firms, do they benefit from like
• 21:48
  global upgrades to AI as a whole as this as AI moves forward and
• 21:53
  you know. I think they do, but it's hard
• 21:56
  to do the attribution as to like how much of the improvement is
• 21:59
  based on our composite, on some model learning on their
• 22:04
  documents and improving for them and how much is from the global
• 22:07
  improvements to AI itself. But essentially, yes, they have
• 22:11
  to, right? Because if those models are
• 22:12
  getting better than our data and we have a pretty large data
• 22:15
  science team that was continually tweaking how these
• 22:17
  different models play in and when there's updates, it might
• 22:20
  affect confidence scores. So we're constantly looking at
• 22:23
  at what's going on in that and we're accessing any third party
• 22:27
  models are all being accessed through, you know, obviously we
• 22:30
  have secure agreements where they're not allowed to train on
• 22:32
  the data and all that. It's all what it it all and all
• 22:35
  that usually goes through your cloud providers like a three AWS
• 22:37
  bedrock, whatever you have like specialized enterprise
• 22:40
  agreements that don't that you can ensure that they are not
• 22:43
  training on data. They also sign HIPAA BA as with
• 22:46
  you and all those sorts of things to because we have to
• 22:49
  technically for us, we are HIPAA compliance.
• 22:51
  So we need any vendor we use to give us a BAA and all of that.
• 22:54
  So. AI is being used in a ton of
• 22:56
  different things now, right? We're seeing like Google's Nano
• 22:59
  Banana and like VO and like people doing crazy shit with
• 23:02
  like it's insane or two. What is the most bizarre thing
• 23:05
  that you have experienced in with the use of AI?
• 23:09
  Does not have to be related to your company.
• 23:11
  No, no, not at all. And I'm not talking about like
• 23:13
  on pizza or like people with like 6 fingers, right?
• 23:15
  I'm talking about like you like you expected some type of output
• 23:19
  and you got something just like off the wall.
• 23:21
  That's a good question. I was going to go a different
• 23:23
  route with what the craziest because I'm, you know, by the
• 23:26
  way, don't let my like Indian Ness fool you.
• 23:28
  I'm like the least tech savvy person in my entire fucking cut.
• 23:32
  Yeah, yeah, yeah, let's. Be I don't know.
• 23:35
  Anything, let's be clear here. I was, I'm just, I was a lawyer
• 23:38
  like, you know, so my background is I was, I was, I was a lawyer
• 23:41
  in a law firm that had this problem.
• 23:43
  I probably tell you that story because it's how this all came
• 23:45
  about. But to answer your question, you
• 23:47
  know, I think the most bizarre thing about AI to me is just how
• 23:50
  realistic now like the video stuff is getting.
• 23:54
  And I if we're going on an entirely side side quest here,
• 23:58
  it's I think that's insane, But I think there's also some
• 24:01
  positives to it because I think that ideal.
• 24:03
  I have young children, so I think about this stuff a lot
• 24:04
  like social media, how old five and three.
• 24:07
  So so I'm wondering what the world's going to look like like
• 24:10
  when they get older. Like, and I'm hopeful that all
• 24:12
  of this like AI slop, as they call it, are going to make
• 24:15
  people kind of like go back to the real world a little bit like
• 24:19
  in in just social constructs and not spend their entire lives
• 24:22
  like on Instagram looking at you.
• 24:23
  Think that's really going to happen?
• 24:25
  You think people are going to? Like a lot of people are, a lot
• 24:27
  of people are saying that now with all of this sort of AI slop
• 24:30
  coming out, this idea that you don't even know like who's real,
• 24:33
  who's not and all this, that people, young people are
• 24:35
  starting to just get tired of it now.
• 24:37
  There's so much like of the content that they're scrolling
• 24:40
  through is like just AI created, you know, nonsense bait exactly
• 24:45
  that people were starting to like be like, well, it's kind of
• 24:47
  like if y'all had Myspace back in the day, right?
• 24:50
  And Myspace was like, I'm dating myself now, Myspace.
• 24:54
  And then all of a sudden it got invested.
• 24:56
  Do you know what Myspace is? I have never been.
• 24:59
  On Did you? No, no, that wasn't the
• 25:00
  question. Do you know what Myspace?
• 25:02
  Is did you have Myspace? No, I did.
• 25:03
  Matthew, you ever heard? Of it, sorry Myspace was where
• 25:08
  it was at and then all of a sudden the bots and all the fake
• 25:10
  accounts came in and everyone's like this sucks and then went
• 25:13
  off it so maybe something there. Yeah, but Myspace was static.
• 25:16
  Yeah, right. It didn't go through upgrades.
• 25:18
  People had a song and. A background and then you put.
• 25:22
  Custom. I had a badass song, I'm telling
• 25:24
  you. What was your song?
• 25:24
  I don't remember anymore, but I wasn't that bad ass.
• 25:27
  Yeah, I know, I know. It's probably Regina Spector
• 25:32
  quite a bit. I have a 10 year old and an 8
• 25:35
  year old. Yeah, right.
• 25:36
  They both use AI for different purposes.
• 25:39
  My older son is using it to learn how to play chess.
• 25:43
  My younger son is using it to learn how to beat video games.
• 25:47
  Yeah, right. Like, there are ways that this
• 25:50
  stuff can be used that can be very productive, but there's
• 25:52
  also of like, just weird. Those are great use cases for it
• 25:56
  though, right? That's exactly what you want AI
• 25:58
  doing. We did this, it was awesome.
• 26:00
  I, I think I posted a thing of this, but like my, my son has
• 26:03
  been learning how to play chess and I'm like, you know what I
• 26:06
  think Gem and I, I can now do like a live mode and I turned on
• 26:09
  the live camera and I'm like, what move should I make?
• 26:14
  And it's telling me based on like looking at the chess board,
• 26:17
  it can tell like what to do and it's predicting moves and it's.
• 26:22
  It's insane. It's incredible you're.
• 26:24
  Cheating at chess against your 10 year old son he beats.
• 26:26
  Me now all the time. You're using Gemini, Are you
• 26:28
  still lost? I did still lose.
• 26:32
  It's become kind of like a a joke between me and him now.
• 26:35
  Like, I can't beat him anymore. He's practiced so much.
• 26:38
  He's using, he's not just using AI.
• 26:40
  He's like he's watching videos. The world has changed so much.
• 26:44
  Why? I think education is going to be
• 26:46
  completely shifted, right? It has to.
• 26:49
  The expensive computer science degrees that you don't have.
• 26:52
  Yeah, right. Expensive law degree.
• 26:55
  But like, computer science is like it was a discipline.
• 26:58
  People would go to school for many years and they would learn
• 27:01
  how to program. And now, like you can program
• 27:05
  without ever having touched code.
• 27:07
  And there's this this whole cohort of people now who believe
• 27:11
  that like vibe coding and like the future is not Why do I have
• 27:15
  to write code? Machines can write the code.
• 27:17
  Why not just? Tell them vibe code is a
• 27:18
  dangerous. What's it do?
• 27:19
  Yeah, man, I I build stuff all the time, right?
• 27:22
  Like you can get to an end point so much quicker.
• 27:25
  How do you keep? But then I will say the other
• 27:28
  side of that is at least still that there's a bunch of like
• 27:31
  vibe coded apps there that have caused massive, like there was a
• 27:34
  massive breach because like someone vibe coded like an app
• 27:37
  that had everyone's stuff in it and they didn't like do basic
• 27:40
  security stuff. That would have been that, you
• 27:42
  know, in enterprise art. Look.
• 27:44
  Where we're at right now, like you currently did this shit at
• 27:46
  all a year ago and now people are making like incredible,
• 27:51
  incredible things. Like we were talking about some
• 27:53
  like very, very routine process that we're trying to figure out
• 27:56
  how to do. And he's like, OK, so we have a
• 27:58
  team member on our team who all she does is just take the
• 28:01
  projects that we kind of come up with and build them in AI.
• 28:04
  Yeah, it's mostly AI. And he's like, why don't we just
• 28:06
  have her build an agent and have her go through this whole
• 28:08
  process and like we'll have the output.
• 28:10
  The alternative was like, let's get Ava and have them spend like
• 28:13
  hours and hours and hours doing this stuff.
• 28:15
  And it's the same thing with with your guys product.
• 28:17
  Right. Like yeah I like had an idea
• 28:20
  yeah and I asked Gemini how it achieved to give me 3 ideas and
• 28:23
  I go give me this idea and he goes.
• 28:25
  Do you want me to write like the?
• 28:26
  The code for it the code. For it said yes and then after
• 28:29
  you give me the code I said I just shared the link to the chat
• 28:32
  and I sent it over to her. I was like go build it.
• 28:34
  I was like, I don't want to deal with.
• 28:35
  It and how long did it take her? No, this is this morning.
• 28:37
  So we'll see. We'll find.
• 28:39
  Out a very small use case application, not a real
• 28:42
  application, but it was like something that normally I'd have
• 28:44
  to just throw over to Ava to just mainly grind out, yeah.
• 28:47
  You can build an app to do the to do like a tiny point solution
• 28:50
  to do something internally, yeah.
• 28:52
  In the world of AI, right, your solution would be considered a
• 28:56
  point solution. And then you have these like
• 28:57
  more robust, like you're going to do everything with AI.
• 29:00
  Where did the point solutions fit in in the future as we go
• 29:05
  through like these constant iterations of AI becoming more
• 29:07
  and more and more robust and comprehensive and less
• 29:10
  hallucinations? I think that the point solutions
• 29:12
  will always have a place. Now how what AI is being used by
• 29:15
  the point solutions is is a question, right?
• 29:18
  So a gentic, for example, like there are lots of things that
• 29:20
  agents are good at. They can't yet do some of the
• 29:23
  stuff we do, right? But we're looking right now, for
• 29:25
  example, at using them to expand what we do.
• 29:28
  So for example, right now a bunch of documents are still
• 29:30
  coming in through portals, right?
• 29:32
  And people have to like remember credentials and log into these
• 29:35
  portals and download documents. So that's a good use case to
• 29:38
  solve with agents, right? You have an agent that
• 29:40
  understands what the portals are, has the credentials and
• 29:43
  when a document, an e-mail comes in that a document is available
• 29:46
  on a portal, whatever can go in retrieve that document, put in
• 29:49
  the system and process it all through to do everything else
• 29:51
  right They. Didn't even think about that.
• 29:52
  Very good. Yeah, I mean, so, so there's a
• 29:55
  lot of and that feeds into, you know, this idea of capturing all
• 29:58
  documents where they come in and ensuring they get into the right
• 30:01
  place. I think point solutions will
• 30:03
  exist. How much of the other AI they
• 30:06
  incorporate is is the question right?
• 30:08
  Do you think the bigger companies are going to eat the
• 30:10
  smaller companies like you guys? Like do you see like a fit at
• 30:14
  some point where foundation could become a component of a
• 30:18
  larger platform or is it exist on its own and like live in the
• 30:22
  ecosystem gets leveraged as needed?
• 30:24
  It's a good question. I think that in the long tail I,
• 30:29
  I don't know is what I think. I mean, it's moving, so
• 30:31
  everything's moving really fast. So where are things going and a?
• 30:35
  Lot of money's coming in. And a lot of money's coming in,
• 30:37
  right? And so when you say the big
• 30:39
  companies are referring to the big companies in in our space,
• 30:41
  here are the big companies like Open AI and, and.
• 30:44
  I was thinking less about like the, you know, the open eyes of
• 30:47
  the world. I'm thinking more of like the
• 30:49
  ones that exist in our space. Yeah, I do think that they're,
• 30:53
  it's a, it's a good question, right.
• 30:54
  I think that right now for a lot of these things that are more
• 30:58
  workflow and more complex and more individualized by a firm,
• 31:01
  most of them are choosing the partner approach.
• 31:03
  Other companies try to build it and it's much more, it's much
• 31:06
  more complex than it seems, especially when you get into
• 31:08
  after you build it for one customer and you go to the next
• 31:10
  and they're completely different and the next is different.
• 31:12
  Plus the amount of document variation it just the longer you
• 31:15
  do it the better. You get, it's not even just
• 31:17
  document variation, it's, it's workflow and all that.
• 31:19
  Like if you think about like Litify right where you were at
• 31:22
  or a file line, right? They're so customizable, right?
• 31:25
  And so every firm has customized it to their own way.
• 31:28
  So if you build a product and you go, great, I built it for
• 31:30
  this firm, it's awesome to take to this firm or that firm.
• 31:33
  Like it won't work. You have to make it so
• 31:35
  configurable that it can meet the level of customization that
• 31:38
  firms have gone through. We had to figure that out the
• 31:41
  hard way over years. But but I think that's one of
• 31:44
  the barriers to, to doing a lot of this kind of work.
• 31:48
  I think also that right now the companies in our space, right,
• 31:51
  there's a whole universe of what I'll call the sexy things
• 31:53
  they're all trying to do, right? You know, I'm sure you've had on
• 31:56
  your podcast, some of the founders of some of those
• 31:57
  businesses doing all of those crazy sexy things like, you
• 32:01
  know, like, like, even if they're all doing a lot of like,
• 32:05
  you know, what I would call more closely to the legal work,
• 32:08
  right? And I think that the less sexy
• 32:12
  part of all this, that is a lot more of the workflow
• 32:14
  orchestration and things like that is, is a little different
• 32:18
  because it can't be done purely by AI.
• 32:20
  It's a lot more, it's really a combination of engineering,
• 32:23
  basic engineering, because you have to integrate with all these
• 32:25
  systems and everything and AI on top of it to be able to optimize
• 32:29
  these processes of workflows. And it has to in, in our case,
• 32:33
  that one of the trickier parts is being able to bridge the gap
• 32:36
  between systems and ensure that we're meeting customers where
• 32:39
  they're at. I was thinking about too, when
• 32:41
  we were talking, we were talking about Litify and file Line, like
• 32:44
  how much of a nightmare must be for you guys to work with every
• 32:47
  platform and like every customer's incident is different
• 32:52
  and in different stages or processes are followed, not
• 32:55
  followed, data hygiene issues. Dated hygiene issues.
• 32:58
  Full implementation, it has to be like you guys have to get
• 33:01
  like ingrained into the firm. As you understand, your inputs
• 33:04
  into their platform and their platform needs to be like good.
• 33:06
  And we're taking inputs out of their platform because in order
• 33:09
  to match a document to the right matter, right, you need to have
• 33:12
  their, they need to have good data.
• 33:14
  We need their data for what data exists on a matter because
• 33:17
  something comes in to a firm, it doesn't have their matter ID.
• 33:20
  It might be a medical record that has a name and date of
• 33:22
  birth. And they could have four people
• 33:24
  or they could be, they could have a, you know, SSD claim and
• 33:27
  API claim and a word comp claim, right?
• 33:29
  And it might be saying from the court that has a case number and
• 33:31
  that might not be in their system or a claim number or
• 33:33
  these documents all have various like identifiers on them that we
• 33:37
  have to be able to look at and match against what's in their
• 33:40
  system. That's a really tricky thing to
• 33:42
  do that and never get it wrong, right?
• 33:44
  So that's why there's humans in the loot, because you cannot be
• 33:47
  wrong. It's it's easier.
• 33:50
  It's better to have someone to err on the side of having
• 33:52
  someone review to make sure which one of these is it just in
• 33:55
  case. Because if you're wrong and you
• 33:56
  put something in the wrong matter or a closed matter or
• 33:59
  something like that and someone doesn't see it, that's huge,
• 34:02
  right? So you have to be right that
• 34:04
  it's the right thing. And so that part of the mix of
• 34:08
  getting it to the right file is the most critical part, right?
• 34:11
  And then after that, obviously it's it's really important to
• 34:14
  make sure that you have it in the right folder and the right
• 34:16
  name and all that. Getting it to the right file is,
• 34:19
  you know, tantamount is, yeah. And most importantly, make sure
• 34:23
  you don't put it in the wrong file.
• 34:24
  Yeah, exactly. Exactly.
• 34:27
  I have one more question for you before we wrap up here.
• 34:29
  Yeah, we talked to so many different law firm owners and
• 34:33
  when we asked the the question about utilizing AI, sometimes
• 34:37
  the responses, I don't want to be a first mover.
• 34:40
  I don't want to make the decision now because I don't
• 34:42
  want to get bought into the wrong platform or get stuck in
• 34:45
  some ecosystem. I mean, it's it's kind of why
• 34:47
  like all the case management systems, like you kind of waited
• 34:49
  to see which was the one that people adopted and now it's
• 34:51
  filed by and litify. Yeah.
• 34:52
  You're stuck with a big contract, a big implementation.
• 34:54
  You're, then you're once you're in and with it, you're kind of
• 34:57
  stuck. They also don't want to be left
• 34:58
  behind. What's the right move for the
• 35:01
  firm owners that are trepidatious about moving
• 35:03
  forward? With what?
• 35:04
  Whether it's your platform, another platform, or doing
• 35:06
  something custom and making the wrong decision.
• 35:08
  Yeah, well, first I'd say probably don't do anything
• 35:11
  custom. I mean for most there's a bunch
• 35:13
  of companies that are investing heavily to build really great
• 35:16
  technology and and you know, other industries we've seen over
• 35:19
  time this idea of like buy verse build.
• 35:21
  It's almost never the right decision to build.
• 35:23
  And if you build, you will dirt certainly get stuck with
• 35:25
  something that isn't going to upgrade, isn't going to get any
• 35:28
  better and you're going to be stuck with what you get.
• 35:30
  I'll. Even say interject really quick.
• 35:32
  I've said this so many times to John.
• 35:34
  He'll be like, let's just build. I go, are we a software company?
• 35:36
  Yeah. He's like, no, I'm like, the
• 35:38
  world's pay for it. I mean, yeah.
• 35:42
  And conceptually it's the same argument, right.
• 35:45
  So then I'd say, you know, with who you go with.
• 35:48
  I mean, I totally understand the worry.
• 35:50
  It's easier for me to speak up my own company because what we
• 35:53
  do then then others because I'm not entirely sure how the
• 35:55
  contracting works. Like, but for us, like we don't
• 35:58
  lock anyone into long term commitments.
• 35:59
  Like, you know, our contract is a year and then it, you know,
• 36:02
  and then customers, we have very good success pertaining
• 36:06
  customers because, you know, once they're using it, like no
• 36:09
  one thinks about going back to, you know, like you don't think
• 36:13
  about getting rid of your spam filter or like not using your
• 36:15
  GPS and printing out like MapQuest again, like the stuff
• 36:18
  we used to do before technology like made it make sense.
• 36:21
  But I would think that you should go.
• 36:23
  But but you know, I think it's really important to vet your
• 36:27
  vendors the same way you like customer vets you, right, Like,
• 36:30
  you know, customers, they look at Google reviews, right?
• 36:33
  Firms should ask, you know, do you have references?
• 36:36
  Can we talk to any of you not just like case that case studies
• 36:38
  are great, but can we talk to any of your references like you
• 36:40
  have any firms similar size, whatever that are willing to
• 36:44
  talk to us? And you know, luckily we do, but
• 36:47
  there there's stuff like that that I think they can do to
• 36:50
  ensure that they're making the right bet.
• 36:52
  And then for us, there isn't really a lot of anyone else that
• 36:56
  does what we do because it's hard and not sexy.
• 37:00
  We've spent a lot of time perfecting it, I'd say, but not
• 37:04
  perfecting it, but getting pretty good at it.
• 37:05
  I will. Say the the CEO of the AI
• 37:08
  company is here telling us that the right way to make a decision
• 37:13
  is to talk to people in the real world to get references, to have
• 37:15
  real conversations. It's exactly what you were
• 37:17
  talking about, right? Like bringing stuff back to how
• 37:20
  it used to be and, and not relying on asking AI.
• 37:23
  What's the best AI platform for my law firm?
• 37:26
  Yeah, without vetting in real. Life, I think it's it's if I
• 37:29
  could summon that ties back to like looking at your firm,
• 37:32
  figuring out, look, lawyers are trained right.
• 37:35
  We're trained to reason by analogy, right?
• 37:37
  Our jobs and what we were trained to do is to like look at
• 37:40
  a, a fact pattern and find, you know, a different fact pattern
• 37:44
  that's similar. Figure out what happened there,
• 37:47
  what precedent replies there and then apply that to our our fact
• 37:50
  pattern, right? Same kind of thing.
• 37:51
  Look at a firm, figure out where your bottlenecks are, where your
• 37:54
  problems are, what problems you're having that need to be
• 37:57
  solved. And then, OK, what's an
• 37:59
  analogous firm? So find are there any firms that
• 38:03
  had a similar problem, right? How did they solve it?
• 38:06
  Well, there you go. I mean, that's the same way that
• 38:08
  lawyers, you know, reason. And I always tell salespeople
• 38:10
  like you can't sell like to lawyers.
• 38:13
  Like, I mean, you could try, but lawyers don't want to be sold to
• 38:16
  like they're human bullshit detectors.
• 38:19
  So like sitting around trying to like objection handle and sell
• 38:22
  them like these dumb sales tactics.
• 38:24
  Like it's like this, it's nonsense.
• 38:26
  You got to like, basically do that.
• 38:27
  Be like, all right, like let's identify what your actual
• 38:30
  problem is. All right.
• 38:31
  Well, how about we take a look if there's other perms that have
• 38:34
  the problem, we show you how they solved it and and then, you
• 38:36
  know, logic would dictate that this would similarly work for
• 38:40
  you to solve your problem. And and that's I think the way
• 38:42
  you got to do it. I will say I will vouch for the
• 38:45
  Foundation sales team. They are fantastic.
• 38:48
  I we are we are we are slightly biased because Carl and Eddie,
• 38:52
  we got to work with them for a few years over lit up by very
• 38:56
  good dudes. We try to hire only good dudes
• 38:58
  and good gals, you know? This has been fantastic.
• 39:01
  If people want to reach out to you or the company, how do they
• 39:05
  get a hold of you? Yeah, you can reach out to me
• 39:07
  directly at vivekvivek@foundationai.com.
• 39:11
  The company is Foundation ai.com.
• 39:13
  And then you can find us online on Instagram, I'm sure, and
• 39:17
  LinkedIn and all the places and e-mail me directly if you have
• 39:20
  any questions or want to talk about our our software or about
• 39:23
  AI in general and where I think it's going with social media and
• 39:26
  children and stuff like that. This was awesome, great
• 39:29
  conversation. Thank you so much for taking the
• 39:31
  time to join us. Sorry we didn't get to catch you
• 39:33
  at MTMP, but we we made it up. Super excited to have done it
• 39:36
  here. Yeah, no, this is, this is
• 39:37
  awesome. This is great, thank you for
• 39:39
  being on. Could have kept going for
• 39:41
  another hour but you're busy man.
• 39:43
  Thank you for being here. Yeah.
• 39:44
  Thank you guys.