Legal Research That Hit 100% Recall — at a Sixth of the Price
Results at a glance
It found 100% of the relevant law on the bar exam, and beat the $1,200-a-seat incumbent on the exam itself.
1,002,364 court opinions in the database, 46.6 million opinion chunks pre-processed for national coverage, 18.1 million citation links in one graph, 70.2 million federal docket records, 14,889 real jury verdicts extracted — every claim pinned to source law, served from a retrieval index that runs on hardware the firm owns for a few dollars a month.
Challenge
Small firms were paying $1,200 a seat for legal AI that still missed relevant law — and sending privileged client matters off to someone else's servers to do it. It found 100% of the relevant law on the bar exam, and beat the $1,200-a-seat incumbent on the exam itself.
Constraints
- check_circleMust find 100% of the relevant law — anything less is unacceptable for legal advice
- check_circleBeat the $1,200-a-seat incumbent on the exam itself
- check_circleCitation integrity required—no hallucinated cases
- check_circleMulti-state support (IA, FL, TX, GA, NY, CA)
Approach
Most legal AI searches for words and hopes the right statute comes back. I built it the other way: it reads what the client's actually facing — who, where, what, why — and goes straight to the section of law that governs it. No guessing.
Architecture
- 01Instead of guessing, the system sorts each question by where/who/what/why and goes straight to the exact law that applies
- 02Every answer links back to the real statute and the cases that cite it — so a lawyer can check it in seconds
- 03Runs on a computer you own — privileged client matters never leave your office
Implementation
- 01A corpus of 1,002,364 court opinions, so it searches real law instead of inventing it
- 02Extracted 14,889 real jury verdicts, so the case valuation has source-grounded damages data
- 03Built the rules that map a client's situation to the exact statutes that govern it
Results
- check_circleIt found 100% of the relevant law on the bar exam, and beat the $1,200-a-seat incumbent on the exam itself.
- check_circle1,002,364 court opinions in the database, 46.6 million opinion chunks pre-processed for national coverage, 18.1 million citation links in one graph, 70.2 million federal docket records, 14,889 real jury verdicts extracted — every claim pinned to source law, served from a retrieval index that runs on hardware the firm owns for a few dollars a month.
Business impact
- check_circleTwo products off one stack: a consumer intake chat and the per-seat attorney research tool
- check_circleAttorney platform that beat the $1,200-a-seat incumbent on relevant-law recall
- check_circle14,889 real jury verdicts extracted — so a firm prices a case against what juries actually awarded, not a hunch
Technologies
The research is public.
Benchmarks run on machines I own, measured against frontier systems — with the failed approaches reported next to the results. Read it before you take my word for anything.