← back to all projects

Production RAG pipeline for document intelligence

−31% hallucination rate

FastAPI · Qdrant · LangChain · Redis · Docker Compose · RAGAS

Placeholder: expand this write-up with architecture notes, screenshots, and results.

Built a Retrieval-Augmented Generation system for document-based question answering using Qdrant as the vector store.

Evaluated 5 chunking strategies and 3 embedding models across a custom 200-document dataset using MLflow tracking. Reduced hallucination rate by 31% and improved answer relevance by 18% on RAGAS evaluation. Deployed as a containerized FastAPI service with Redis-based caching for retrieval latency.