Lab.
(01) Lab
Things I'm building. Small, working builds that show the method: real extraction, real evals, real cost numbers. New entries ship here as they're ready.
- 001In progress
Clause extraction with an eval harness
Upload a contract, extract key clauses, and see accuracy, confidence and cost per document scored against a ground-truth set. It's the evaluation discipline behind a production legal AI platform, in miniature.
- LLM extraction
- Ground-truth evals
- LLM-as-judge
- Cost per document
Have an AI bet you can't size?
30 minutes. No pitch deck. If I'm not the right fit, I'll say so.