Lexata: using GPT-3 search endpoint to find legal rules

Lexata is a legal tech start-up focussed on capital markets regulations. Capital markets rules are highly technical and even experts struggle to perform research efficiently. Lexata’s goal is to make the rules more accessible by letting users input plain language queries and presenting the search results (drawn from Lexata’s database) in order of relevance. I’ve done some testing in the playground and I think it’s a terrific use case for the search endpoint. My biggest concern, however, is whether it will be economic to build a commercial product using GPT-3; if each search is too expensive, it won’t work. Love to hear anybody’s thoughts. At present, Lexata’s site is powered by Relevanssi search. Thanks.


I really like this idea, especially since when most people study Government in high-school or college, they usually end up learning about logrolling and omnibus bills that are used to sneak in policies through some form of obscurity since whoever is good with that is able to further provision bills for their own benefit and it can be quite hard to catch them by reading stacks and stacks of paper.

Since’s there’s purposely a whole lot of junk to get away with this stuff, having GPT-3 quickly analyze and point out these instances would greatly reduce bad actors in that realm.

I would look into this, but I have little experience with how logrolling works on a granular scale to feel like I could personally assist, but I don’t see why it’s not doable!

1 Like

@lmccallum If you provide numbers in terms of projected number of users, and usage statistics, should be able to estimate API costs. Most likely you’ll be using a finetuned ada or babbage model, which is 0.004 - 0.006 / 1000 tokens, though I’m not 100% sure how tokens get counted for search. If this doesn’t scale, then there are open-source alternatives that might fit your use-case better.

Probably best to prototype and validate using gpt3 search endpoint, then gradually move to custom solutions.

I worked at LexisNexis, and although it was a long time ago, I am pretty confident that most of the following is still true.

  1. Legal research services are expensive, and big law customers are willing to pay a LOT–hundreds or thousands of dollars per associate per month–since the results influence the outcome of billion-dollar litigation and are billed back to clients. that said, clients manage costs very carefully these days, so traceability & justification are important. Everything about pricing is even more true for capital markets, since dollars involved are vast.

  2. Law firms like all different kinds of billing arrangements – subscription, transactional, and hourly. Back in my day, the complicated accounting backends were a major constraint on development.

  3. Legal researchers are very demanding. I have a couple of expert friends who could provide you with insight on a consulting basis.

  4. Query formulation is important, but don’t neglect the results. While lawyers probably don’t want you to paraphrase the law for them, they may find it very helpful for you to propose a summary they can show to clients.

  5. At a few cents per transaction, cost is not likely to be your blocker.


Thanks @NimbleBooksLLC, all of your points are very well taken.

Thanks @asabet. Yes, I’m building an MVP with the search endpoint while thinking about future customizations.


This is great, thank you for sharing.

I would like to get in touch to start developing a similar use case but for a different audience. Can you help on how to get started?

Sure, you can e-mail me at lmccallum@lexata.ca if you want to discuss. Is your background legal, technical or both?

Thanks, will get in touch via email. My background is mainly technical.