Wexler Bags $5.3m – Interview With CEO, Gregory Mostyn – Artificial Lawyer

Wexler, a startup focused on leveraging AI for complex litigation workflows, has gained $5.3m in a Seed funding round led by Pear VC, with participation from Seedcamp, Myriad Venture Partners, and The LegalTech Fund. (See In-depth AL Interview with CEO, Gregory Mostyn, below).

The funding comes as the company launches Wexler Real-Time, a legal fact-checking feature that flags false testimony, and adds to its ability to build chronologies, query case materials, and surface key inconsistencies. Clients now include: Clifford Chance, HSF Kramer, Goodwin Procter, Burges Salmon, and Addleshaw Goddard.

Kathleen Estreich, Partner at Pear VC, which led the investment round, commented: ‘At Pear, we love to partner with highly ambitious founders who are tackling big problems with unique tech. Greg and the wexler.ai team are building what’s quickly becoming essential infrastructure for litigation. They combine real-time fact-checking with workflows designed for how lawyers actually win cases, becoming essential for all litigators.’

And now the In-depth AL Interview with CEO, Gregory Mostyn. Press Play to watch inside the page. There is a full AI transcript below.

–

You can find more about Wexler here.

—

Legal Innovators Conferences in London and New York – November ’25

If you’d like to stay ahead of the legal AI curve then come along to Legal Innovators New York, Nov 19 + 20 and also, Legal Innovators UK – Nov 4 + 5 + 6, where the brightest minds will be sharing their insights on where we are now and where we are heading. 

Legal Innovators UK arrives first, with: Law Firm Day on Nov 4th, then Inhouse Day, on the 5th, and then our new Litigation Day on the 6th. 

Both events, as always, are organised by the Cosmonauts team! 

Please get in contact with them if you’d like to take part.

—

AI Transcript of AL In-depth Interview:

Yeah, so we’ve raised $5.3 million, led by Pear VC, who are a Bay Area based seed fund, one of the most eminent seed funds in San Francisco, alongside Seed Camp, who have a number of legal tech investments already and are, know, the partner there, Tom is an ex lawyer and have been very ….. one of having a bunch of massive investments Revolut, and wise, UiPath, etc. And then the Legal Tech Fund, I’m sure will be familiar to your listeners.

Richard Tromans (01:24.35)

Okay, fantastic, fantastic. And this is seed, pre-seed, A, what is it?

Gregory Mostyn (01:30.682)

It’s sort of a late seed, I suppose. You know, we’re a bit ahead in terms of ARR and traction of most seed companies. yeah, I wanted to raise now sort of expand to the US, you know, need capital to, to, to develop and grow in such a fast moving market. So decided this is an opportune time to sort of put down some roots, expand the team, expand the product offering and really with a focus on tackling the biggest market, which is the US without, know, losing focus on our existing

European customers as well as those in the APAC region, but really to win in the US you need capital and that’s why we’ve made the decision to fundraise.

Richard Tromans (02:08.38)

Yeah, yeah, yeah. And I remember, I remember, think I was probably one of the first people you chatted to, at least in the media, calling up and saying, I’ve got this new legal tech product and you’ve come a long way very fast. So when did you just remind everybody, when did you launch?

Gregory Mostyn (02:13.36)

Yeah.

Gregory Mostyn (02:22.608)

Yeah, so we launched publicly last April, which I think was also around the time when we spoke, maybe a little bit before that. Essentially, before that, we were sort of building in stealth. think it’s an interesting, it’s an interesting, you know, the traditional thinking in startups is sort of build a scrappy product and release it, you know, an MVP style situation, but that just doesn’t work in this market, especially when you’re building something as nuanced and specialized as litigation, it really needs to be good. And so we spent a lot of time in customer development, speaking to our customers, understanding exactly what kind of pain points they wanted us to solve. And, you know, we took on a lot of advice, as you may know, from last time, my father, who’s a high court judge, was very involved in that initial build. And, yeah, it started off as a tool just for building chronologies, essentially extracting the facts in litigation, and it’s become this fact intelligence platform we call it, which essentially is really solving the world’s most complex disputes by compiling the factual record looking for inconsistencies. And now with this new release, actually, the new paradigm, which is also ingesting audio. So transcribing depositions interviews. Yeah.

Richard Tromans (03:29.438)

Hmm. Oh, well, it’s very, it’s very timely that we’re doing this in this interview in this way, then it’s perfect. You can, you can try and apply Wexler to this interview. think that would be interesting. So, just, just to remind people, so where, you know, where, where is the moat? Where is the added value? People might say playing devil’s advocate. Oh, well, yes, but there are other tools that can do this. What, where’s, where’s Wexler sort of forging its own path?

Gregory Mostyn (03:36.752)

Yeah, exactly. I could do it.

Gregory Mostyn (03:58.738)

I think there’s a couple of key differentiators. The soundbite I always say is we’re a scalpel, not a Swiss Army knife. So we go really deep into specific workflows. We build custom workflows where the human is kept in the loop, which operate at a fact level rather than document level. So in comparison to eDiscovery, which essentially might prepare documents for human review, Wexler actually reads the documents like a human does, picks out the key factual insights, i.e. the events inside those documents, even if it’s buried on page 993 hidden in a footnote of an attachment, we’ll pull that out, enrich it, establish why it’s important to the issues in the case, and allow the user to verify and sort of manipulate and use that data going forward. So it’s a question of reading documents at a fact level, the scale with which we can operate. Wexler can actually ingest 500,000 documents with that level of detail, which is pretty, I think, without parallel for a GenAI native platform. And

Yeah, it’s about going really deep into specialized workflows and building things with the nuances, complexities for dispute resolution, where the fact patterns are so varied, not just as an afterthought applied to something which really has been trained for transactional work and litigation to the crowbar afterwards. So yeah, I suppose those are the kind of thing where I’m thinking.

Richard Tromans (05:12.382)

Yeah, that sounds very useful. I mean, one thing and I’ll ask it because obviously lawyers will ask it anyway, because it’s something they need to ask, which is, on a minute. So this is using generative AI. This is all about facts. How do we know the facts are facts? And it hasn’t imagined that Brad Pitt is somehow involved in this case and he’s not.

Gregory Mostyn (05:18.609)

Yeah.

Gregory Mostyn (05:26.81)

Yeah. Yeah.

Gregory Mostyn (05:32.789)

Yeah. Well, this is where you have to build an intelligent UI, which essentially allows the lawyer to verify the fact that it’s been extracted within milliseconds. So we actually have a document viewer, which can rifle through a 10,000 page document in basically under a second to scroll to the correct page. So the lawyer can really quickly see exactly the paragraph, the section, the sentence that it’s been taken from essentially. So you have a doc viewer with a highlight. It shows you exactly where it’s been taken from.

shows you the thinking of why it’s been applied to the case, i.e. why it’s relevant. And so, every single time you see anything in Wexler, it always has a source and a citation. It’s built without making it really onerous and time-consuming to do so. It’s built around the human verification of AI output. And that’s what it has to be. Because as you say, you need to have that expert human to be able to verify the output and develop it.

Richard Tromans (06:24.718)

But broadly you don’t need to do that with every one because I guess if you had to check every single fact it would not completely reduce the efficiency.

Gregory Mostyn (06:30.586)

Yeah, of course. I think the thing is, like, we have extensive evals, we’ve done a huge amount of testing, and it’s broadly extremely accurate. Like, we can’t say it’s completely accurate, because of course, there are nuances. There’s no fabrication to the point where it won’t, like, hallucinate the kind of word de jour, but essentially, it won’t hallucinate that Brad Pitt is in the case, you know, you’re getting really stuck in complex legal language, you need to correct, you need to be comfortable and sort of verify that at the point of producing the final piece of work product. So it’s not when you’re sort of filtering through reams and reams of facts, it’s when you get to that final piece. It’s not just about correcting it for accuracy, but it’s about writing it in your own style. It’s about developing it. It’s about making a first draft into something that you have put your own, you know, Richard Truman’s stamp on, which is exactly the way it should be. And that process actually helps the lawyers get better at understanding the case. So it’s not just about correcting and verifying this human in the loop part of the process actually makes the humans become experts. So it’s not just, the AI told me so. They can actually speak to their clients or in court, you know, with confidence.

Richard Tromans (07:36.03)

And can you customize it? you say, you know, I’ve got one of my junior associates, or if you’re a barrister, you know, one of the junior barristers to go through it using this system, but I’d like to look at it for a particular lens. I want to look at it in relation to X or Y legal concept in particular, or if you’re a judge, for example, and you want to put your particular frame over this data. Can you do that?

Gregory Mostyn (07:51.43)

Yeah. Yeah. Yeah. Yeah.

Yeah, absolutely. So basically, when you upload the set of documents, you can give it a list of issues, the axes by which you want to analyze the documents, what it will then do, it will analyze all the facts, it will classify them as relevant, maybe relevant or irrelevant to those issues. And it will also tag each fact if it’s appropriate to do so with those issues. So breach of contract, theft of IP, whatever it is, it will tag based on that list of issues.

You can also upload the claim form or the complaint and our system will automatically extract the issues if you choose it to do that way. Or as you say, you might just want to analyze it with reference to one specific issue because that’s, say it’s a public inquiry and there are 10 different, you know, defendants and you only care about one specific thing, you can analyze it with reference to that specific issue. So yeah, absolutely.

Richard Tromans (08:44.158)

And are there any, obviously without mentioning any names, aside from family members, are there any judges using this?

Gregory Mostyn (08:47.474)

We’ve had a lot of judges extensively testing the product. We’re currently trialing with a and getting really good feedback with the Barristers Chambers as well. So we are actually expanding into the bar, which is a great, you know, vote of confidence given their exacting standards. So yeah, I think absolutely we’ve had a lot of using it testing in the UK and the US. And we are commercializing into the bar, which is interesting, because I don’t think

know, there’s not been much legal AI that’s had traction in the bar so far.

Richard Tromans (09:20.254)

Oh no, it’s hard. mean, we, well, we could do a whole, we could do a whole chat on why it’s difficult to get tech into the bar. But as, people know, or if they don’t, if they’re not in the UK, we have a split profession, solicitors and barristers. Barristers generally work in chambers. Uh, each individual barrister is truly independent and you could have two barristers sitting in the same room, working on the same case, but on different sides of that matter, which a lot of people find quite extraordinary. But because of that, it’s quite difficult to have large scale tech implementation.

Gregory Mostyn (09:23.922)

Yeah.

Gregory Mostyn (09:33.052)

Mm-hmm.

Gregory Mostyn (09:37.233)

Mm-hmm.

against each other. Yeah.

Gregory Mostyn (09:46.171)

Exactly.

Richard Tromans (09:50.23)

certainly which involve data because, for obvious reasons. So yeah, it’s very, very interesting. It’s great that you’re doing that. On the US side, I think. So you’ll have an office. Do you have an office already in America?

Gregory Mostyn (09:50.746)

Yeah. Yeah.

Gregory Mostyn (09:59.42)

Yeah.

We don’t have an office yet, but we are regularly in the US, you know, going out for trade shows, visiting customers, closing deals. Part of this funding is going to be used to hire employees in the US basically set up an office over there. And as we expand through the funding rounds, we’ll be developing our imprint even more. You know, it can be managed UK to East Coast. So I’m heading out there every couple of months to visit customers spending time in person. Better than a lot of tech companies going to San Fran, you know, that is a lot more challenging. But given most of the big law firms, the DC

Chicago, New York, Boston, it’s manageable.

Richard Tromans (10:33.724)

Yeah, and of course, we hope to see you at Legal Innovators New York in November, in several weeks. And also California next June, which is it may not be the legal center of the world, but it’s definitely the legal AI center. So final point really, where do you go from here? So how many people, how many staff are you these days?

Gregory Mostyn (10:37.956)

Yes, yes, for sure. I’ve got a call actually. Definitely.

Gregory Mostyn (10:48.24)

No, I think that’s true.

Gregory Mostyn (10:54.384)

Yeah. How many staff? How many staff did you say? So with about 12 at the moment and expanding rapidly. So as well as advisors who you’ll know, like Tara Waters, who’s a former advisor to the business and various other people who operate on a fractional basis. yeah, somewhere between 12 and 20, depending on which people we’ve got data labeling and people involved in various different projects we’re running. So yeah, sort of lean but mighty. And I think actually, it’s amazing what you can now do with AI and be so as you’ll know,

Richard Tromans (11:22.899)

Mmm.

Gregory Mostyn (11:23.762)

be so much more productive from a technical perspective. I saw Ross McNairn’s post about his new release where all of the code was written by AI. And we’re definitely leveraging that to increase productivity in the team without having to hire extensively, which is a really interesting time to be building a startup. So yeah, in terms of where we go, it’s just like doubling down on this vision of becoming the platform for litigation AI, expanding the scale to millions of documents from 500,000. We’re already getting there soon, as well as more custom use cases for

analyzing depositions, know, reviewing testimony, preparing for trial, specializing in different jurisdictions, potentially, and maybe even in the future, applying the law to the facts, because obviously, that’s the next step. Once you’ve got the factual matrix is to then apply the law to that. But for now, really focused on just building more functionality around cracking the facts in the case, building a winning case strategy, second level review after eDiscovery. And yeah, all early case assessment.

basically anywhere in the in the dispute that isn’t the core of the disclosure discovery process. So yeah.

Richard Tromans (12:25.406)

Gotcha. Just a very last question, just for the LLM geeks out there, what Gen.A.I. systems are you using? I presume you’re using the major large language models, are you using anything else?

Gregory Mostyn (12:30.919)

Yeah.

Yeah.

Gregory Mostyn (12:36.646)

Yeah. Yeah, so it’s model agnostic, basically daisy chaining together multiple different models in a LLM pipeline, which I think is, you know, how, you know, on neuro symbolic approach, if you want to buzzword, which basically is how we’ve, you know, made the LLMs do the finite amount of reasoning they can per step. So we have our kind of worker models, which tend to be GPT-4-0, Gemini 2.5 Flash, then we have the higher reasoning models, which might be Claude 4, Gemini 2.5 Pro.

or GPT-5, which we are using, although actually we found that the evals for Claude 4 for the production of the answers we find to be a lot more concise than GPT-5, interestingly, but we are still using it in some areas. And so yeah, when you upload documents, it goes through this kind of stitched together pipeline and then has all of these documents and facts extracted in an index, which you then can go back and retrieve and query again using a graph rag system. So yeah, it’s a very

technically complex and sort of deep pipeline if you like. yeah, that’s how we’ve, we’ve kind of built this thing where facts are the oil that powers the product, know, facts are the gold.

Richard Tromans (13:44.937)

Fantastic, fantastic, Greg. Good stuff and good luck.