Sardine's real-world fraud test: Jev hits 93% recall vs 62% for LLM classification
jamdac · x · 2026-10-29
Sardine CEO Soups Ranjan shared results from testing TypeSafe AI's System One model Jev on a real fraud incident, ahead of a joint event with a16z.
- Jev classified best at 93% recall; velocity rules proposed by Sardine's Data Analyst Agent followed at 90%; an LLM making the classification decision trailed at 62%.
- After the test, Sardine began incorporating Jev into its platform.
- The event will cover fraud ring detection, onboarding risk signals, and automated KYB checks, with TypeSafe CTO Erik Gafni joining a panel moderated by a16z's Yoko Li.
More from Venture
- Dev thanks 'shitty' GitHub Actions as his side project rakes in Stripe payments — aniketmaurya · 2026-10-29
- Modal Hits $1B Annualized Run Rate and Raises Series D Funding — charles_irl · 2026-10-29
- Old vs new users: AI natives skip Google, sites need both SEO and AI traffic — gefei55 · 2026-10-29
- Anthropic revenue up ~65x in 18 months as valuation multiple falls to 21x — PEEEPOOpeeepoo · 2026-10-29
- Gegei's Overseas Tool Suite to Launch English Version, Targeting Foreign Paid Users — gefei55 · 2026-10-29
- $2M seed at 20% dilution means every $100k burned costs 1% of your company — dunkhippo33 · 2026-10-29