How legal AI Legora hit $100M ARR in 18 months: its retrieval stack, explained
DevinLewtan · x · 2026-11-13
turbopuffer published a case study: legal tech firm Legora reached $100M ARR in 18 months, serving 130,000+ legal professionals across 80+ markets, with 4B+ legal docs indexed (2B+ for search per the linked page) for research, drafting and review.
Architecture evolution: they started on self-hosted Elasticsearch, but large firms contractually demanded per-client storage isolation—separate clusters per customer were unworkable. They moved to pgvector + DiskANN on Azure, inheriting a database-per-tenant pattern via Postgres plus a proxy, which solved isolation but introduced new issues. The final migration to turbopuffer delivered 100ms p90 latency and 98% recall@10 while combining legal-grade isolation, high recall and low latency. Their documents team handles extraction, parsing, chunking, embedding and indexing, with the core goal of keeping agents from wasting tokens on wrong context—at their scale, inference budgets are supply-constrained, so token efficiency defines the production ceiling.
More from Venture
- Hacker House Alumni Launch Mission Street Capital, a VC Betting on SF Talent Clusters — jamdac · 2026-11-13
- Companies led by Indian-origin CEOs hit $15T combined valuation, AI drives 71% — nordicinst · 2026-11-13
- Bloomberg: SpaceX and Tesla could announce merger as soon as December — rohanpaul_ai · 2026-11-13
- Before buying $50K AI SEO tools, spend $1-2K setting up BigQuery first — gaganghotra_ · 2026-11-13
- Cerebras bear flips positive at $83, citing <0.5% CS-3 wafer decommission rate — firstadopter · 2026-11-13
- At least 42 of Bessemer's AI 100 have Nvidia on their cap tables — aleximm · 2026-11-13