Evidence-Grounded Interview Feedback Copilot
Surfaces what interviewers actually said, organized by competency. Every claim is grounded in verbatim text from the original debrief. No recommendation is ever produced — that stays with the hiring committee.
Analysis Pipeline
Step 1
Extract
Baseline keyword matching or Claude tool_use
Step 2
Verify
100% citation validity gate — every span is checked
Step 3
Analyze
Coverage map + disagreement detection across interviewers
Step 4
Synthesize
Evidence-grounded report, no recommendation language
Step 5
Review
Human reviewer adds notes and approves before sharing
Non-Negotiable Design Constraints
- ✓No hire/no-hire recommendation — the tool surfaces evidence, humans decide
- ✓Every model-generated claim cites a verbatim source span from the debrief
- ✓Pydantic validation gates all LLM output — invalid output is rejected, not silently accepted
- ✓Human review is a required step, not an optional add-on
Tech Stack
FastAPIPydantic v2Python 3.11Claude tool_useNext.js 14TypeScriptTailwind CSS