From the Legal Trenches to Tech

Nick Coleman [00:00:00]: I didn't realize that this was a profession where you helped claimants or people that are applying for disability to advocate for them through this process. And so it was a very interesting introduction because this firm handles so many cases at one time, I was literally handling 300 cases a year.

Demetrios [00:00:29]: I'm really excited to talk to you because of your position, how you are the subject matter expert who has turned into a founder that is using AI to leverage their product and company. And when I say leverage, I mean that you're almost an AI native company is what I think the best term would be. So let's start with just setting the scene of what exactly are you doing, what is the company and how are you attacking the space that you're in?

Nick Coleman [00:01:05]: Yeah. So in a nutshell, we are providing a suite of tools, leveraging large language models to help streamline the pain points of Social Security disability attorneys. Social Security disability attorneys advocate for people, claimants, they that are applying for disability benefits because they no longer can work. It's a welfare kind of safety net in the United States. I'm pretty sure that most European countries have these. You essentially pay into a trust fund. So if you get hurt, you have a safety net. And so if you that pays you a monthly benefit.

Nick Coleman [00:01:44]: As with any government process, it's a bureaucratic nightmare. And so attorneys are often needed to get involved, to help advocate for clients. And so what we're doing because of the bureaucratic nightmare, with documentation, all types of evidence that's needed to help advocate for them, we are using large language models to streamline a lot of those processes.

Demetrios [00:02:08]: Okay, we'll talk about the exact processes that you're streamlining. But to set the scene more for me, as someone who is not and has not ever filed for any claims, basically I get injured and I am a W2 employee or I. I'm a employee or even I'm a business owner. I have my own llc and I can then go to the government and say I can't work for the next six months or a year. I would like to get paid while I'm not working.

Nick Coleman [00:02:41]: Is.

Demetrios [00:02:42]: Is that what I'm understanding?

Nick Coleman [00:02:43]: No, it's generally you need to establish that you have a disabling condition that's going to prevent you from working for 12 months, consecutive 12 months, and what they call severe impairments. And so it can be a range of conditions from mental to physical, whether it's from major depressive disorder or to having a nerve root compression in your spine that's going to prevent you from working for 12 months. The arduous part of the process is getting the evidence to establish why you can't work. And that's what attorneys are doing. It's collecting all the evidence from the different people involved in the process, from either your doctors or getting information from your employers to explain why you no longer can either do your past work, the job that you're currently in, or even in the past, or why you couldn't do an easier job. Walmart greeter. You know, you think about like, why couldn't you do a very easy, simple job? And you have to establish it for a 12 month period. And you are, your payments are based on what you've paid in Social Security taxes.

Nick Coleman [00:03:55]: So in the United States. So when you get withholding out of your paycheck, that's how they determine what your monthly benefit would be.

Demetrios [00:04:04]: All right, and now the attorney gets involved. Because if I fill out all the paperwork and then I get rejected, that's where the attorney gets involved. Or is it when I'm filling out the paperwork?

Nick Coleman [00:04:15]: There's different attorneys get. Depending on attorney strategy and business model, oftentimes attorneys will pick it up initially, and other times attorneys don't want to get involved till later on. Because there's a process where you have essentially four opportunities to establish your disability. And then after the fourth shot, the agency or the Social Security administration says, hey, you haven't established it. What happens is most people end up applying individually online, where they fill out an online application, provide the medical documentation, and then if they are denied, they call it this first step. Then they will call an attorney to help them collect the necessary evidence and kind of navigate this like, like I said, this bureaucratic nightmare. Because it's a lot of convoluted regulations that don't really make any sense. And, and attorneys have to kind of, they, they do this all the time so they understand what's the most streamlined process to, to effectively get you those benefits.

Nick Coleman [00:05:18]: And that we're kind of the translator of, of, of this, this regulatory framework that.

Demetrios [00:05:27]: And this is where it gets interesting because that, if I understand correctly, is what you used to do.

Nick Coleman [00:05:33]: Yes. And so I started out my journey. I've been practicing for 12 years now. When I got out of law school, I graduated 2011. It was right after the recession. It was a joke. You had a better opportunity becoming a barista than you did getting a job at a law firm. And so honestly, at the beginning, I was substitute teaching.

Nick Coleman [00:05:55]: I went one week from reading Cat in the Hat to kindergartners to accepting a job at a Social Security disability practice. At the time, I didn't know anything about Social Security disability other than my mom had received benefits and that you had a Social Security number. I didn't realize that this was a profession where you helped claimants or people that are applying for disability to advocate for them through this process. And so it was a very interesting introduction because this firm handles so many cases at one time. I was literally handling 300 cases a year.

Demetrios [00:06:37]: Wow.

Nick Coleman [00:06:39]: 40,000 miles on the road, going to hearings all over Arkansas and the region. And then at the same time, they expected me to do what is a more technical part of this job, which is appellate drafting. Think about the real deep research type drafting, essentially auditing the Social Security administration. So what if a case is denied? Then that portion of the job is to identify all of the errors there that have been made when the Social Security administration reaches a determination, because at each stage they have to provide a notice and explain why you were denied. And so I had to do both at the same time, which is a very. To ride out of the gate. Not knowing anything about it was pretty scary.

Demetrios [00:07:31]: Yeah.

Nick Coleman [00:07:31]: And big time imposter syndrome. And so obviously with, with, with being a lawyer, you have to worry about malpractice ethics. Can I effectively advocate for all these people at the same time? And that's where that was kind of the catalyst of how I started leveraging technology with this job.

Demetrios [00:07:56]: Yeah. So now take me into the journey that you have had into tech.

Nick Coleman [00:08:01]: Yeah. So like I said, I was scared to death when I got into this practice area. But I've always been a computer nerd, Always trying to find ways to automate processes. And so like I was saying, a lot of what this advocacy job is, is developing the record, essentially, and getting that evidence needed to establish why someone's disabled. And so what is that? What is that evidence? That's. That's the medical evidence. And oftentimes you're dealing with thousands of pages of medical evidence and you're trying to find a needle in the haystack. And so at the time before we had all, you know, the, the AI that we have today, you are digging through thousands of pages of medical evidence and trying to dig out insights within that evidence, because oftentimes there is a regulation that has some mapping to a clinical finding.

Demetrios [00:09:03]: So wait, so it's not just as easy as me saying, hey, my back hurts. I broke my back. I can't sit anymore. And the doctor signed off on that?

Nick Coleman [00:09:14]: Not at all. It's sometimes you can have three docs treating doctors, saying, explaining why you're disabled, and the agency will still deny you.

Demetrios [00:09:24]: What?

Nick Coleman [00:09:25]: Yes.

Demetrios [00:09:28]: Wow.

Nick Coleman [00:09:28]: So half your job as an advocate is not just trying to find that needle in the haystack through thousands of pages of evidence. It's honestly developing the record enough so if it is denied that you have an opportunity to appeal it, to give that person another chance so they can, for instance, if a judge denies it, what I. The way I'd strategically handle these cases, I present every. I try to get everything that I can to establish they're disabled. But I also have to think, what kind of traps can I set in the record that will help me get this case appealed and sent back? So it's always. You're always looking at it both ways because there's only. There's. Like I said, there's four stages.

Nick Coleman [00:10:17]: The first stage, only about 85% of people get or get denied. The second one higher. And then they. And they actually. Actually, after that, you go in front of a judge, and then there's only a 40% chance that you get approved at that stage. So to be a really effective advocate, you have to look at it real time, say, how can I get them? How can I win this case now? And how can I set this up effectively so if they do get denied that I can set a record up to get this overturned and give them another shot at winning? Wow.

Demetrios [00:10:52]: Well, you said something in here that just automatically made my. A light bulb in my head go off, which is finding a needle in a haystack. And I'm like, oh, wait. They literally had a test for large language model context windows about that. How have you started to incorporate different AI processes into this? Because it feels like it's ripe for many different areas.

Nick Coleman [00:11:20]: 100%. And so what we're doing. And in the United States, they use what they call ICD10 codes. And I'm sure Europe has a probably similar system of how they bill diagnosis codes. Right. The Social Security disability framework is built on ICD10 codes. As far as what I was talking about, severe impairments. And so what we're trying to do, um, there are severe impairments, and it gets really nuanced.

Nick Coleman [00:11:49]: Then there's like, severe severe impairments. Like, okay. And it's called the listing of impairment. And the whole thing that I was very good at when I was doing this work in the weeds, 300 cases a year, was finding that needle in the haystack and using essentially Adobe and plugins to create repositories of Words to match the regulations. You could almost think of a technical sense named entity recognition, I guess. And so that's what we were doing. And so what I would be able to do is flag words to help me find the things that I know I needed to establish to win that case. And at the time I was like, man, it would be awesome if you could have this automated system that did all this, because even with the plugins, it was still an arduous process.

Nick Coleman [00:12:34]: Yeah, but it was better than. I joke. We all just like control find, right. To go through a 2000 record. It was better than. Than that. And so this was kind of the catalyst of why I created LexMed. And we actually, one of my best friends that I went to law school, we initially came up with this idea.

Nick Coleman [00:12:53]: And even before OpenAI, we were looking at different solutions through AWS, like medical comprehend and things like that. But it still wasn't good enough. And so when OpenAI came out in November 2022, the things that I was trying to do, I could start doing that. And so that's. And I went ham, essentially using the logic that I've been doing manually and creating these mapping techniques and building out the logic. Because a lot when I talk about logic with Social Security, it's very much like a system where you're using a sequential process with decision tree, conditional decision trees. And so using that and leveraging my understanding of that logic and then putting it into which things like a mega prompt with large language models. And that's how I segued into leveraging AI and starting this company.

Demetrios [00:13:57]: And so it is from the get go, if we go back to these different work streams that you had, one is you're trying to get someone there, you're trying to get someone not refused, and then you're also appealing refusals. And at each step of the way, you're building a case. And I am assuming you're looking at what the doctor has given you and you're building the case from various doctors who. And you're also putting into the case these back doors like you were talking about for the repeal.

Nick Coleman [00:14:35]: Exactly. Tra. Essentially setting up. Yeah. Setting up booby traps for appeal. And so if. Then that's what effective attorneys do in this practice area is have looking at it both ways to always set it up for success in case things go wrong. Right.

Nick Coleman [00:14:53]: And so that's the way I set it up and to throw kind of a wrench in this thing when I'm talking about these four stages of appeals that you go through the first two stages, they're going to deny your case based on what a contracted doctor has said about your client. So the government has their own doctors that will review the medical evidence as they get it. And then they're obviously very conservative because they can't give everybody benefits. And so those doctors who have not physically examine your client are determining how their impairments impact their ability to work. And ultimately those opinions become what you have to overcome. And so that's like the adverse evidence. So they have, you have four state government doctors that say you can't work. And so it's the attorney's job to develop the record, whether it's through treating medical providers opinions or, or diagnostic testing, MRIs, ultrasounds, whatever you can.

Nick Coleman [00:15:55]: Because they, to establish it, the framework requires both medical opinions and objective findings. They do use subjective complaints like why, how do you feel? But it has to be, there has to be a direct connection of how your allegations are aligned with the objective findings in the record. So there's a lot of, a lot of moving pieces.

Demetrios [00:16:18]: Is this where that mapping and those numbers come in? Where you say, for example, you have a spine injury and that maps to a certain number and the doctors for the state are looking at that and saying there's not enough evidence here that that number that you're claiming you have is real.

Nick Coleman [00:16:40]: You hit the nail on the head. And so this is what I bet it was very effective at over the years. They have what they call listing of impairments and one of them is that you fit under the musculoskeletal listing. So they have a category for each body part. And so one thing that I was effective with, it was 1.04 regulation Degenerative disc disease with nerve root compression. And so it has all of these qualifying factors that all need to be met within a 12 month period. And the assumption is if you have all of these factors, there's not a question of whether you're going to work. The assumption is you're so messed up you can't work.

Nick Coleman [00:17:21]: But they're very difficult to meet. And so for instance, and I'll use this example of finding that needle in the haystack, 1.04 at the time would require diagnostic imaging. So an MRI establishing that you actually have nerve recompression in your spine. Then you have to have a physical examination that shows you have limited range of motion. Then you have sensory loss, reflex loss, muscle loss. And then there's another test called a straight leg raised test that indicates nerve recompression in a 2000 page record. You might not exhibit all of these symptoms in one visit. And so you are tasked with piecing essentially a puzzle together to make your argument that they meet these requirements.

Nick Coleman [00:18:09]: And this is where I was trying to. I could create a repository of words which essentially was control F on steroids. Right. To actually match those. So they're giving me a cheat sheet to see if they met these requirements. But with large language models, it changed the whole game as far as how I did that, because large language models can distinguish between synonyms in the medical record. And this is where it's big. So where you could use, I guess you could say regular expressions and things like that.

Nick Coleman [00:18:44]: It was very arduous task because medical evidence doesn't use the same word to describe one thing. And your question about the back impairment is a great example. Think about how many ways that you can describe reduced range of motion in someone's back. Yeah, Limited. Reduced. And then there's a. There's. In medical.

Nick Coleman [00:19:05]: Different medical framework record frameworks use different words and there might be 20 ways to describe the same thing. And that's where large language models really are helpful with that. They can distinguish between very similar words that mean the same thing under the regulatory framework.

Demetrios [00:19:26]: So first of all, I love your use of the word arduous. That's a great vocab word that I'm going to try and incorporate into my vocabulary. And second of all, I think I'm getting a better picture here now of you've got these records, these medical records that are books. It's like Iliad and the Odyssey. You've got all of these conditions that need to be met in order for someone to successfully pass and get their claims. And you are trying to match the two and build up this case that says, look, judging on when they went to the doctor in January 2022, and then they got this MRI and then they did this and you show all these things that make it. So it is very clear they have done everything for whatever it was at 1.04 that you were talking about, that all they jump through the hoops or they tick all of the boxes. And you are now kicking off that type of matching with a large language model, parsing through all of the records, all of the medical records.

Demetrios [00:20:40]: And then the large language model, I'm assuming, is also parsing through all of the numbers and all of these codes. So whether it's 0.1.

Nick Coleman [00:20:52]: Yeah. So you, you. What we've been focusing on and we're get. We're about to Build out our chart vision platform fairly soon is leveraging what's available now as far as using RAG to essentially put a database of this regulatory framework that's hundreds of pages to help match these clinical findings to the ground. You know, ground truth is like the regulatory framework and then match it with the medical record. That changes dynamically because people are constantly treating. Right. And things change where they might not have met a particular listing in January of 2025.

Nick Coleman [00:21:33]: They now have a new diagnostic test that shows that they do have nerve root compression. And so that's how we're, that's how we're kind of leveraging the large language models with that.

Demetrios [00:21:46]: And I can see a world where you're also, as new legislation comes out or as these different tags get updated, that is getting updated too. So now you not only need an MRI and to have these different criteria being met, but you also, because they updated something in the database from the state side, you need this extra box to be ticked. And the large language model can parse out that in these large bills that come through.

Nick Coleman [00:22:18]: Again, you're hitting the nail on the head. So what I said used to with 1.04a they made it harder in 21. Yeah. And so now they found out that too many people were getting approved under 1.04 a. So they've essentially added the requirement that you have to be completely compromised in your upper arms so that you effectively can't use them or you require a walker. So it's like, okay, you have. Now they've essentially made it as okay, you have degenerative disc disease and you require a walker. Okay.

Nick Coleman [00:22:49]: And you have, you have an amputation and you require a walker. I mean, I'm not really exaggerating. That's essentially how difficult they've made it. So now we've have to find different strategies to get around that. But that's what, but that's what's so great about technology and these new tools because the technology allows us to pivot with the. When there are fundamental changes in the framework. And so we're going to be doing that with what our chart vision project. And we're already have done it with our, now of our product that we're doing with our transcription services where we're actually doing auditing on hearing testimony where they've changed some regulations just recently.

Demetrios [00:23:37]: Talk to me more about that.

Nick Coleman [00:23:39]: Yeah, so when I, like I said this, this LexMed Legal Medical, that was my, this has been a passion project of mine. But to get to market early and because of the technical feats that we needed to deal with the, the security apparatus, the amount of money that we need to do to get everything and do it right. With the medical analysis we started out and we are, right now we are transcribing hearings. And so there's, I know there's a lot of different transcription services out there, but what we're doing is actually adding that subject matter expert flair to it and actually helping identify and translate words that are often misinterpreted with off the shelf solutions by using my context and understanding of what the way things should be said and who the speakers should be. And so we are using my data that I have where we've got actual hearing audio and have the human transcripts that act as the ground truth. And so we're using that to find patterns in our automatic speech recognition model using regular expressions to swap out words. And then we're actually adding speaker labeling based on who is found in the hearing. And so we released that last year and we've taken it to the next step of actually using kind of my auditing logic with the way that I review decisions.

Nick Coleman [00:25:17]: So, you know, we talked about how we're building a case by using the medical record to help establish disability. This is on the flip side. I'm using my logic to look at how the Social Security Administration messes everything up to find errors. And we're using large language models to actually parse the transcript from the automatic speech recognition model and identify all of the errors that have happened with expert testimony in the hearing.

Demetrios [00:25:51]: And now this is to presumably, if I'm understanding it correctly, understand the new updated laws or requirements or this is more general and it's for any hearing.

Nick Coleman [00:26:05]: This is the same thing except a different use case with as, as far, far as what kind of evidence we're using. And so during an administrative hearing, this is where they go where they've been denied twice. The claimant's got to go in front of a judge when they go testify. The way they evaluate whether they're disabled is whether they can work or not. And so they hire an vocational expert who's an expert in jobs to testify whether they can do other work. Even if they can't do their past job, can they do other work? And they frequently mess this up. And so this is where the agency uses jobs that don't even exist anymore to find that people can, that aren't, that they're not disabled. They, there's literally jobs that haven't really existed in the national economy in 30 years.

Nick Coleman [00:27:00]: That these vocational experts say they can do. So nut sorter, unskilled surveillance system auditor. I joke. A blockbuster rental clerk. You know, it's essentially that, right? And so that's not an actual one, but it might as well be a dinosaur wrangler, a model T mechanic, you know, just on and on. But there's a regulatory framework with this on, on with the vocational evidence. And so what we're doing is actually having the large language models audit their testimony using this regulatory framework and databases. Like we're using a SQLite database that has information on all these jobs.

Nick Coleman [00:27:44]: And we're having the. We've just started using MCP server and functions to actually reconcile whether the job that this expert testified actually exists, whether that person could actually perform that job. For instance, so you like with a back impairment, that person can't bend at the waist more than occasionally. Well, the job that the person that the vocational expert said they could perform requires frequent stooping. It's something as little as that. But attorneys miss out on this because there's so much going on in the hearing. What we're doing is kind of playing that safety net. And so if it is, something did go wrong, they can use this cheat sheet essentially to file something with the court right after and said, hey, this testimony is bogus.

Nick Coleman [00:28:34]: It. You can't use this to deny the case. Or if they do deny it, what you've done is, is you've stabbed. You've. You've set the record up for appeal so that they can effectively appeal that case so they can get another bite at the apple if it is denied.

Demetrios [00:28:53]: And what you're bringing here is the speed at which someone can understand when something is bogus.

Nick Coleman [00:29:02]: Yes, exactly. The thing with Social Security disability practices is it's all about efficiency. You think of an attorney that gets paid on the hourly rate. That's not the way Social Security disability practices. It's contingency. And so think about going back to me working with 300 different clients in the year. You're constantly trying to prioritize how you should spend your time. It's just the economics of it.

Nick Coleman [00:29:28]: And so you are often left with, man, do I even put in the work to spend on a case that has a 50% chance of getting sent back. I've got five hearings tomorrow. I don't have time for this. Right. That's just the way. That's just the reality of the situation. What we want to do, and the whole reason I want to build this is to make. I don't want people to have to make that decision.

Nick Coleman [00:29:56]: I want everybody to put. I want every attorney to be able to be that super advocate so they can effectively advocate for every claimant equally. And that's what I see with AI as being like, the great equalizer is that I want everybody to be as big of an audit nerd as me and be happy, be able to effectively advocate and give their clients the best chance at winning. And that's ultimately what, in a nutshell, what we're trying to do with all these tools.

Demetrios [00:30:31]: It's gotta be demoralizing when you have that situation where you're thinking how to prioritize and you have to let certain ones go. Because even though you know in your heart that you, this person should be getting their claims pushed through, for the amount of work that it's going to take you to get those pushed through, you just don't have the time. And the probability that it will actually go through it is not high. So you have to prioritize something else. And that seems like now if you are able to transcribe it in the moment, and then I'm guessing that what you're giving that attorney is like red flags that they can automatically look at.

Nick Coleman [00:31:20]: Exactly. We're giving them the cheat sheet. Here's the deal. Most of hearing most attorneys aren't listening to this audio in the first place. One, it sucks. Nobody likes to listen to their own voice. And these hearings you can imagine are boring as sin. So it's like, so they don't even do it in the first place.

Nick Coleman [00:31:37]: And so what we're doing is like, hey, look, you don't have to listen to these horrible hearings. We know you're not doing it anyways because when you do, you're just looking for these gold nuggets, these red flag flags, right? And since you're not even doing that, you're probably not appealing and developing the record anyways, right? Because you've got five hearings on Friday. You don't have time to sit there and scrub and rewind and do all this to find these things that you don't even know what to do with if you do find them. And so what we're doing is just making it so easy that you don't have to listen to the audio. They don't have to do the research to, to determine whether this expert testified, whether, you know, most people are going to know a dinosaur wrangler job doesn't exist anymore. But hey, maybe a surveillance system monitor that job might exist. Let's look, let's have the system red flag it and explain why it doesn't. And that's, that's kind of the brilliance with this is that again.

Nick Coleman [00:32:34]: Yeah, we're giving you that opportunity so you don't have to create this mental triage is what it is. Right. Well, Kathy's got fibromyalgia. You know, she's really hurting. But I've got another client that actually has nerve root compression and has a walker. I don't have time to spend, I don't have two hours, three hours to spend on Kathy's case to build the record when I need to focus on Joe's back. Right. We want to remove that from the equation and that's ultimately in a.

Nick Coleman [00:33:06]: That's what we want to do and that's, that's kind of what we are doing with the suite of tools. It's not just obviously it is streamlining the, this boring work, digging through medical records, finding that needle in the haystack, listening to these cringe worthy audio recordings that nobody wants to listen to. But the end goal with this is that I want everybody to be that zealous, effective advocate that can adequately represent these claimants and give them the best chance at getting their claims approved. Because so many people are denied for arbitrary reasons, unfair system, broken system. And I don't want it to be because the attorney had to prioritize whether they took on more dedicated more time to Joe's messed up back over Kathy's, you know, fibromyalgia.

Demetrios [00:34:02]: So that is powerful. What in your mind comes next, man?

Nick Coleman [00:34:10]: No, I just, I wanted to say I'm really excited. We are actually going to be expanding this, these, this tool set. We're doing a big fundraise and investments of strategic law firm partners that I'm excited they're going to help scale this company and I'm really excited about what we can do in this space. And I've got two awesome dedicated developers give them a shout out. Philip Cannon and Nick Burka. And they have been awesome in the whole product development cycle and taking my kind of crazy ideas and actually putting them in to the logic to make it work. And it's really impressive. I'm very excited and I'm really excited you allowed me to come on today.

Nick Coleman [00:34:57]: It's been awesome. I think this all is built on the great equalizer that AI can be. You know, everybody talks about, you know, the legal industry is very scared of this technology but the way I look at it is it's democratizing justice. And democratizing justice is a big thing when you're dealing with governments that have rigorous, sometimes arbitrary regulations with people without the means to be able to adequately advocate for themselves. And so that's what I think is very powerful about this technology. And if leveraged appropriately and ethically, I think that it can be a overall good for society.