Alarie: Reliable legal research synthesis was impossible before LLMs
Ben Alarie · He "kind of" had PMF for 8 years—until, after a rebuild, he raised $100M | Ben Alarie, Founder of... · PMF Show · Oct 20, 2025 · at 19:45
Ben Alarie, founder and CEO of tax AI platform Blue J, explains why earlier supervised learning approaches failed at comprehensive legal research compared to RAG paired with LLMs.
0:00 / 0:38exact quote · 38.0s
▶ Watch the full episode on YouTube → 720p mp4 · rendered on demand · StarZero watermark
“The big move was retrieval augmented generation, where you kind of run a search. So you have a master corpus of all of the relevant tax research materials. You run an intelligent search, and you've got, like, really good vector embeddings, and you're chunking all this stuff and indexing it in just the right way. You can find the relevant research resources, and then you can Synthesize them and generate a very good synthesis of all of the relevant materials and produce an answer. But you just could not do that reliably prior to the large language models having sufficient natural language, understanding synthesis capabilities.”
quote is from the automated transcript, cleaned for reading: filler sounds and stutters are removed, nothing is rephrased. names can be misheard (the analysis reads context, assessments check outside sources). how →
More from Ben Alarie
Alarie: Horizontal LLMs generate flawed tax answers from stale web sources
“More commonly, there will be some of the materials that those systems find even on the open web are at a date anachronistic. The law has changed. There's been some statutory amendment or there's some new case law or, you know, some law firm wrote some guidance…”
Alarie: Blue J raised a $122M Series D in July 2025
“We raised a Series D round, and that closed in July of 2025. And it was a hundred and twenty two million USD.”
Alarie: Tax partners won't build startups due to high opportunity cost
“It might be a tax law partner in a big firm, but that's so extraordinarily unlikely because they already have clients that they're servicing. They're part of a partnership. They're kind of bought in. They kind of have dedicated their lives to that path. It cou…”
Alarie: Pure AI engineers won't solve tax research without deep domain mastery
“Might it be somebody who's more kind of pure tech, like an AI or machine learning background? And again, I thought, well, probably not because they would need to know a lot more tax law than they do. It'd take them a decade to catch up to where I was in my und…”
Alarie: Blue J's early ML models predicted court rulings with over 90% accuracy
“So we built a bunch of these different predictive models, which performed extremely well, like better than we thought they could perform. And so we're predicting with better than 90% accuracy, how courts would resolve these kinds of cases.”
Alarie: RAG shifts legal AI effort from retraining models to data curation
“The brilliant thing about leveraging retrieval augmented generation is we can just curate the data. And if we're confident about the currency of the tax database and the content in the database, and we're confident about our retrieval algorithms, our Prompt en…”