A small studio for careful data.
Delta Evals is built around a simple idea: AI teams need expert judgment and real-world footage — not just more volume.
Who runs Delta Evals
PhD-level researchers in large language models · Published methodology · Toronto-based
Delta Evals is led and managed by PhD-level researchers working on large language models and evaluation. The same researchers design the guidelines, calibrate the labelers, and quality-review every batch we deliver — nothing is farmed out to a general crowd.
Our published methodology covers multi-agent labeling, soft-label distributions, model calibration, and evaluation design, with prior work in AI trust and safety and applied research at industry and government labs.
Tell us what you’re working on.
We usually start with a short pilot for labeling or video.