Add Gemini backend and stabilize single-worker deployment

[AyushAnand413]

This PR improves document QA reliability by switching the generation layer to support Gemini and stabilizing deployment behavior on Hugging Face Spaces.

Changes included:

• added Gemini backend support with a shared client factory
• updated supervisor and response generation flow to use configurable LLM backends
• added google-genai dependency
• changed Gunicorn to a single worker to avoid in-memory document state issues across requests
• kept the existing Hugging Face backend as a fallback option

Environment updates:

• LLM_BACKEND=gemini
• GEMINI_MODEL=gemini-2.5-flash
• LLM_INTENT_BACKEND=gemini
• GEMINI_INTENT_MODEL=gemini-2.5-flash
• GEMINI_API_KEY added as a secret

Local verification:

• Gemini smoke test succeeded
• prompt-level generation test succeeded

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
ragify-ai	Ready	Preview, Comment	Mar 23, 2026 6:56pm