Prototype: upload a resume + target job title, 3 parallel Claude agents (Title, Content,
Credentials) score it, an arbitrator reconciles their flags into one score. If the score is
below threshold, it asks 2-3 targeted follow-up questions (answer by text or voice — voice is
transcribed locally with faster-whisper), then re-scores. Reports are exportable as a static
HTML page. All storage is local JSON files in data/sessions/.
cd oHR
cp backend/.env.example backend/.env
# edit backend/.env and paste your ANTHROPIC_API_KEY
venv/bin/uvicorn main:app --app-dir backend --port 8420 --reload
Then open http://localhost:8420
- First voice answer will be slow (faster-whisper downloads the "base" model on first use).
- No database — sessions are JSON files under
data/sessions/<uuid>.json. - Threshold for triggering follow-up questions is
THRESHOLD = 6inbackend/main.py.
