AI Career Coach

Keyless demo

demo corpus · deterministic demo embeddings · canned generation

You are chatting over the résumé of Avery Patel, a fictional persona. The gates are the same TypeScript the live path runs — conformal OOD abstention and human-review routing — over a deterministic demo embedding space. Answers are canned or verbatim excerpts (no model call), so what you are evaluating here is the gating, not generative quality.

Have keys configured? Use the live coach.

Pick a scripted question above, or ask anything.

On-résumé questions get a grounded answer. Off-résumé questions hit the calibrated abstention gate instead of a made-up answer — that's the point.

Demo honesty: the persona is fictional, the embedding space is a deterministic hashed-lexical construction (not a learned model), and answers are canned or verbatim excerpts — but the abstention threshold is genuinely conformal-calibrated on this space and every gate decision you see is made by the production gate code at request time.