“System One” models like Jev are fast general classifiers. Classifiers have existed since 1958, but they have to be trained for specific tasks: if you build a…

What this adds A new example, examples/system_one_distillation, that tests the idea from Sean Goedecke's System One models can train their own replacements: use a capable decision model to ...

Judge recorded sessions with TypeSafe's jev model by asking your own yes/no, choice, and score questions, one evaluation result per question.

Agent skills for designing and building durable AI agent workflows with Kitaru - zenml-io/kitaru-skills

Summaries and timestamps for every talk on the AI Engineer YouTube channel.

Replay-based evals for AI agents: your production traces, re-run against your next change. Built for agents that write into a system of record, where testing in production would create phantom bookings and duplicate claims. Import the runs your agent already made, turn what your team notices into an evaluator, then compare two experiment runs to see what a new model or prompt would have done. Open source, self-hosted, built by the ZenML team.

In this walkthrough, Alex takes Kitaru from the high-level idea to ...

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

LightOnOCR-2-1B: a lightweight high-performance end-to-end OCR model family