AI / ML2025-01
Enterprise RAG & Knowledge Base Platform
The company had millions of internal documents spread across 10+ data sources — employees were wasting hours searching for information that existed somewhere but was impossible to find. I built a RAG platform that indexes all of it with a scalable ingestion pipeline handling 10k+ PDFs, using Qdrant for vector storage and LangChain for retrieval orchestration.
Key Highlights
- The key decision was keeping everything on-premise with Llama 3 to maintain data sovereignty
- It delivers context-aware answers in under 1 second with 95% retrieval accuracy
- Started as a prototype for one team, now runs organization-wide
Technology Stack
AI
Llama 3LangChainOpenAI
Infrastructure
QdrantMeilisearchUnstructured.ioStreamlitDocker
Core
FastAPIPython