OpenAI (and Gemini fallback)

LLM providers for structured JSON outputs in the hiring flow.

Used for

• Resume parsing — suggest candidate fields from extracted text (PDF/DOCX text extraction is local, not sent as binary to the model)
• Interview reports — scores, skills, recommendation from transcript and context (OpenAI only on this path)

Extraction can fall back to Gemini’s OpenAI-compatible endpoint when the primary call fails with retryable errors.

Auth (env vars)

• OPENAI_API_KEY — primary
• GEMINI_API_KEY — optional fallback for extraction
• GEMINI_FALLBACK_MODEL — fallback model name (default in config if unset)

Notes

• Keys optional at boot; parse/report fail at call time if required key is missing.
• Do not log full prompts or completions in production; log ids and token metadata only.