Enterprise RAG Chatbot

Responsibilities & Achievements

• Production Deployment & Infrastructure: Deployed and operated the platform on an Ubuntu server using Docker Compose to orchestrate 8 containerized services (FastAPI, Celery workers, PostgreSQL 18, Redis 8, Qdrant vector DB, RustFS/S3, Traefik, 9Router), with health checks, persistent volumes, and automatic restarts.

• Reverse Proxy & Networking: Configured Nginx as the production reverse proxy with SSL termination, proxy buffering disabled for SSE streaming, and host-based routing across multiple services. Set up Traefik in staging environments with Docker label-based routing.

• CI/CD Pipeline: Designed GitHub Actions workflows with 4 parallel jobs — guardrails (line ending enforcement, docs-first policy), backend linting (Black, Flake8), frontend TypeScript verification, and Gitleaks secret scanning on every push and PR.

• Docker Optimization: Wrote multi-stage Dockerfiles with BuildKit cache mounts for pip and npm, reducing image build times by ~60%. Implemented non-root container users and Docker secrets for credential management.

• RAG Backend & Retrieval: Built a hybrid retrieval engine combining Qdrant dense vectors, BM25 sparse search, and PostgreSQL hierarchical graphs with adaptive reranking — achieving 88.6% Hit@3 accuracy across 6,700+ lines of ERP documentation.

• Caching & Cost Optimization: Implemented multi-tiered caching (L1 hash + L2 Redis semantic cache), reducing warm-query latency from ~3.5s to under 50ms and cutting LLM API token costs by over 40%.