Legora has about a hundred engineers, compared with Lauritzen's earlier goal of staying at twenty.
“Uh, and we're a hundred engineers now, and it feels like we're way too small.”
Person
Jacob Lauritzen
1 episodes · 24 quoted claims
Legora began as a ChatGPT-based tool with a Swedish law search function.
“The way everything started was ChatGPT with, uh, Swedish laws search tool in it. That was literally the entire everything.”
Legora's feature for bulk-reviewing many documents is called tabular view, a term now used across the legal AI industry.
“So one of our features is called tabular view, which is, um, now a name that's being used in the industry.”
Jacob Lauritzen went through Y Combinator in 2023, and Legora went through it in 2024.
“Yeah, I was in YC 23, Legora was 24.”
Sub-agent delegation, used for fan-out retrieval queries, was difficult for Legora's early agents but now works well.
“Sub agent delegation was one of them that we've gotten working really well now for, like, fan out type retrieval queries works extremely well for us now.”
Legora's evals team has about three engineers and three lawyers, supported by a pool of legal engineers.
“there's like three engineers and like three lawyers right now. And then we have a pool of what we call legal engineers.”
A citation problem in Legora's agent went from about 7% to about 99% success after a single new model release.
“we had one problem where, like, it wouldn't cite properly, and so it would be difficult to follow. And it was like, next model release solved. It went from like, 7% to like ninety nine.”
Lauritzen says the partnership model exists because a professional services firm's main capital is its people.
“Because the only type of capital that these companies have is human capital.”
Legora was originally named Leia when its founder applied to Y Combinator.
“reading Max's YC application, uh, and like, helping him getting into YC, get Leia, as it was called back then, into YC.”
Lauritzen says some legal work, especially in the US, is moving from hourly billing to project fees.
“That a bunch of stuff is actually being priced on like a project fee and not on an hours fee.”
Lauritzen says Legora's tabular view supports up to about a million cells.
“we do up to a million cells, I think,”
Legora hosts its infrastructure on Microsoft Azure, mainly because early law-firm customers distrusted other cloud providers.
“all of our infrastructure's on Azure, um, primarily for this reason.”
Law firms treated Microsoft as an approved vendor, which forced Legora to host on Azure, though attitudes have since changed.
“So, like Microsoft is an approved vendor in their mind. And so we had to host everything on on Azure, and that's changed over the past few years.”
Lauritzen says an early Google Gemini model performs very well for transcription and speaker attribution.
“Gemini, one of the early Geminis, is, uh, yeah, extremely good for certain types of, um, uh, transcription and speaker attribution.”
Legora keeps very few vendors on its data processing agreement to control where customer data goes.
“like our data processing agreement has very few vendors on it”
Legora's tabular view first runs a sample of ten rows so users can check results before launching a full run.
“our tab of the view actually will run the first ten rows”
Legora acquired a UK company called Wexler that does fact intelligence for litigation.
“made another acquisition in the UK called Wexler, and they're doing fact intelligence, which is for litigation mostly.”
Legora's first forward deployed engineer was also its first New York hire, a former Y Combinator founder, hired roughly nine months before the interview.
“So we hired our first forward deployed engineer, that was our first hire in New York, also YC, uh, founder, nine months ago.”
At Legora, legal engineers configure the product without writing code, while forward deployed engineers write production code.
“Legal engineers configure the product as a power user would. They don't write code, and a forward deployed engineer writes code into production.”
Legora began offering APIs only around six months before the interview.
“we started our APIs like six months ago or something.”
Legora had not fine-tuned any models until recently, when more capable open-weight models made it worthwhile for some tasks.
“so we haven't fine-tuned any models up until now. I think now, in the past few months, open weight models have become extremely capable.”
Lauritzen cites legal research as a task where a post-trained open-weight model could beat frontier models on cost and latency.
“Legal research is a great example. It's like a you have to make, it's a it's a fan-out retrieval problem”
A Legora engineer worked with the Turbopuffer team to get multi-vector search built for Legora's search needs.
“He actually talked to the TurboPerf team to get them to build Multivector so that we could do these types of things.”
Legora's citation system lets users click through to highlighted source passages, even through layers such as tabular review cells.
“make sure that what it says always is citable somewhere, even multiple layers down.”