Sudapedia Instance

"The core architecture follows a micro‑service style where the FastAPI application acts as the sole API gateway, routing requests to the Gemini LLM for natural language understanding and to MySQL for structured data retrieval. The API is stateless, with JWT‑style bearer tokens issued on POST /auth/login and validated on each protected route. Conversation state is maintained client‑side via a session_id stored in sessionStorage, and the backend returns a thread_id that the client re‑sends to keep the context consistent across requests.

Data ingestion is handled by the extract_all.py script, which parses PDFs into structured JSON and populates the two MySQL databases (DB1 for product metadata and DB2 for SudaHub data). The Gemini client is lazily instantiated to avoid cold‑start latency on Cloud Run, and a retry wrapper with exponential backoff ensures resilience against transient LLM failures.

Security guardrails include CORS permissive for development but can be tightened in production, Bearer token authentication, and environment‑based configuration to keep API keys and database credentials out of source control. All chat logs are persisted in the chatbot_chat_logs table, enabling auditability and future analytics.

The build topology is straightforward: a Dockerfile builds a lightweight image, a Procfile ensures Cloud Run listens on 0.0.0.0:$PORT, and the requirements.txt pins dependencies such as fastapi, uvicorn, google-generativeai, and mysql-connector-python. Continuous deployment can be achieved via Cloud Build triggers that push to Cloud Run on each commit to the main branch.

Overall, the system demonstrates a clean separation of concerns: the API layer, the LLM layer, the data layer, and the frontend UI, all orchestrated through simple REST endpoints and a minimal configuration footprint."