RAG and tatters

Threadbare implementation of a RAG toolkit:

• document chunking
  • fixed length
  • Recursive Character Split (RCS)
  • document specific splitting
  • semantic splitting
• document retrieval (exhaustive search for a one-liner "vector database")
• Approximate Nearest Neighbor (ANN) with Navigable Small World (NSW) for K-Nearest Neighbor (KNN).

Building NSW

KNN