RAG architecture · 5 November 2026
RAG query planning: why a story is not a search query
A fictional tenancy deposit story shows how query decomposition turns a narrative into focused searches, checks evidence coverage and supports cited answers. Includes the workflow, failure modes and a fair evaluation method.
Read the article
RAG architecture · 24 September 2026
Chatbot, RAG, or RAG with MCP tools: which one does your document problem actually need?
One finance question put through all three systems: where each answer comes from, what each can and cannot do, where the risk moves when tools are added, and how to tell which one a proposal is really offering.
Read the article
OCR · 17 September 2026
Invoice OCR is the easy part. Validation is what lets finance trust the result.
A worked example on one scanned invoice: the fields OCR read, the nine checks applied, the one low-confidence field that fails, and what the reviewer sees. Plus the checks every invoice pipeline should run and how to evaluate one.
Read the article
Legal RAG · 14 September 2026
Why legal RAG needs hybrid search: reciprocal rank fusion, explained with lawyers' queries
Lawyers ask two kinds of question that no single retrieval method answers well. How hybrid retrieval fixes it, how reciprocal rank fusion merges the results, and what fusion still cannot do: reranking, chunking, temporal validity and permissions.
Read the article