AI / RAG

RAG Architecture for Enterprise Use Cases

Chunking strategies, vector store trade-offs, and the evaluation loop that keeps retrieval honest in production.

·

10 min read

Retrieval is a product problem, not a search problem

Most retrieval failures are not caused by the vector database. They come from chunk boundaries that split meaning, metadata that never made it into the index, and an evaluation loop that nobody runs after launch.

Design patterns worth copying

Chunk on document structure rather than a fixed token count. Keep a metadata filter in front of the vector search so tenant and permission scoping happen before ranking. Store the source span with every answer so a reviewer can verify it in seconds.

Then measure. A small labelled set of real questions will tell you more than any benchmark.