Hey PH 👋 Most legal AI hallucinates citations. Lawyers know this. They don't use it. I built Pariksha to fix that with two things: 1. A marketplace where you hire legal AI agents per task across 8 jurisdictions (India, Singapore, UAE-DIFC, EW, Korea, EU, US-Delaware, US-Generalist) 2. An open-source benchmark every agent is publicly scored against. 40 questions, citation-verified against primary sources. Apache 2.0. github.com/Aritra003/pariksha-benchmark We publish the failures too : including a hallucination we caught on EU DMA Article 3(2) during benchmark dev. That find made it into the public archive. Three things I'd love feedback on: → Lawyers: does per-task pricing beat per-seat for occasional use? → Benchmark builders: what's missing from our methodology paper? → Anyone: what's a legal AI failure mode we haven't tested yet? Honest critique is worth more than upvotes. Will respond to every comment today.

Pariksha by NyayaMitra

Hireable legal AI agents, scored against an open benchmark

Hireable legal AI agents, scored against an open benchmark