I binge-watched Inventing Anna on Netflix this week. Real-life defense attorney Todd Spodek (played by Arian Moayed) spends time buried under a real truckload of discovery documents, hunting for the ones that matter.
That's the same grind eDiscovery teams run on every real matter — and it's exactly what a new AI model that dropped this month can help with. It's called Jev, from a startup called TypeSafe AI.
TypeSafe AI's Jev is a “decision model”: feed it a document and a set of yes/no or multiple-choice questions, and it returns a calibrated answer for each one, in parallel, for a fraction of what an LLM costs.
If the speed and cost claims hold up, cheap large-scale classification could genuinely reshape what “reasonable” looks like in discovery — the difference between one lawyer buried in a truck full of boxes and a model clearing that same truck in an afternoon.
Here's what doesn't change either way: whether it's an LLM or a model like Jev on the other end, a privileged document still has to leave the premises before a model can judge it. Cortega routes every request under the firm's own data policy, keeps privileged work on local or on-prem models, and logs everything that crosses the boundary, so “what touched this document and where did it go” always has an answer.
Fast and cheap is a real step forward. It's still not the same claim as governed.
Full breakdown of what Jev actually is (and isn't) — Part 2 is here.