TypeSafe AI and Sardine: Fighting Fraud with Jev

Most fraud and compliance teams share the same constraint. They have to make a decision in a few hundred milliseconds, and they don't have the labeled data to train a model that can make it.
In this fireside chat hosted by TypeSafe AI and Sardine, Soups Ranjan, CEO and co-founder of Sardine, and Eric Gafni, CTO and co-founder of TypSafe AI, talk about what happens when you put a fast, calibrated decision model inside a workflow. They share what Sardine learned testing Typesafe's Jev on name matching, KYB, and a real fraud ring, and how to build systems whose decisions you can inspect and explain.
This session covers:
- One-shot decisions for name-to-email matching and KYB, with no hand-tuning
- Why a written SOP has to come before any agent, and how human review on sanctions alerts tapered from 100% to 10%
- How stateless models change your architecture, and why micro-decisions are easier to audit than reasoning traces
- How Jev and a frontier LLM handled the same fraud ring
- How AI shopping agents and fake job applicants give themselves away
- Audience Q&A on drift, interpretability, and rules overriding pretrained knowledge


By submitting this form, you agree that Sardine may contact you about our products and services. For more information, review our privacy policy.