Overview#
Built a modular Retrieval Augmented Generation platform that ingests local PDFs and recursively crawls documentation websites before chunking, embedding, and indexing content for semantic search.
Implemented semantic retrieval using LangChain, Gemini embeddings, and Qdrant Cloud with session isolated vector collections for multi-tenant document indexing.
Added query intent routing, prompt injection detection, toxicity filtering, and groundedness evaluation to reduce hallucinations and reject unsafe requests.
Built a React frontend with streaming responses, source citations, and an execution trace displaying pipeline stages, latency, and token usage for debugging.
Designed a decoupled FastAPI backend exposing REST APIs for document ingestion, retrieval, indexing, and chat workflows following a modular service architecture.
Automated testing and deployment with GitHub Actions, deploying the backend to Google Cloud Run and the frontend to Firebase Hosting.
Tech Stack#
Python
Core language
FastAPI
REST API backend
React
Frontend UI
Vite
Build tool
Tailwind CSS
Styling
LangChain
RAG orchestration
Gemini Embeddings
Vector embeddings
Groq
Fast LLM inference
Qdrant Cloud
Vector database
REST APIs
Architecture
Pytest
Testing framework
Docker
Containerization
Google Cloud Run
Backend hosting
Firebase Hosting
Frontend hosting
GitHub Actions
Automated pipelines
LLM-as-Judge (Evals)
Evaluation
CI/CD
Continuous integration