Personal project · Previously listed as CLOSER
Levelhead
A technical-interview mentor connecting practice to answers and project evidence. Application code controls question selection, findings cite answer spans, and append-only events support replay. Evaluator accuracy has not been benchmarked.
- TypeScript
- Fastify
- PostgreSQL
Ask.
Inspect.
Replay.
- Plan a question
- Cite the answer
- Replay the events
The problem
An interview mentor needs to connect practice to actual answers and project code, while making its findings reviewable rather than presenting an unvalidated readiness score.
Implementation decisions
- Separate the generative interview conductor from the evaluator; select question actions in application code.
- Attach findings to quoted answer spans and allow the evaluator to abstain when evidence is insufficient.
- Record plans, questions, answers and findings as append-only events, with replay rebuilding the session view.
Evidence and limits
The public repository is now opx0/CLOSURE; its README names the product Levelhead. The README, question-selection policy and event-replay implementation were reviewed on 21 October 2026. The benchmark has not been run and the labelled dataset is empty according to that README. Evaluator accuracy and role validity are not established. The visual here is a system illustration, not a product screenshot.