Overview
Built a production-ready RAG System with FAISS Vector Database. The system uses FAISS to store the embeddings of the documents and then uses the embeddings along with clustering to search the documents.
Key Results
Identified optimal k=10 clusters using elbow method and silhouette analysis
Reduced search space by 8x through k means pre-clustering before FAISS indexing
Reduced storage requirements by 35% through efficient numpy array indexing in FAISS.
Achieved <500ms retrieval latency for FAISS vector search across 10K+ document embeddings.
Tech Stack
PythonFAISSHugging Face