Staff · 02
Arun Vale
Partner, Applied AI
A model you can score on Friday, not applaud on Tuesday.
Arun asks, before anyone opens a notebook, what “good” looks like for a model that will sit in front of customers. He designs evaluation as part of the product: labelled sets, judges, rule checks, and a number the release manager can read.
He has shipped retrieval systems that honour the same permissions as the source of truth, and copilots that degrade to a known path instead of inventing policy.
Qualifications
- Applied ML in production, B2B product companies
- Evaluation harnesses wired into CI
- Permission-aware retrieval and operational limits
Works on
- Evaluation datasets built with the people who know the domain
- RAG systems whose access control matches the source of truth
- Tracing, fallbacks, and spend — the parts a demo leaves out