08-12-2026, 09:25 PM
A retrieval-augmented system is more than uploading documents to a chatbot. Start with clean source documents, stable identifiers, and access controls. Chunk text by meaningful sections, preserve metadata, retrieve a small relevant set, and require citations in the answer.
Evaluate with real user questions: did the right document retrieve, did the answer stay grounded, and did it cite the relevant section? Include an “I do not know” behaviour when sources are absent.
What source type would you use first for an internal knowledge assistant?
Evaluate with real user questions: did the right document retrieve, did the answer stay grounded, and did it cite the relevant section? Include an “I do not know” behaviour when sources are absent.
What source type would you use first for an internal knowledge assistant?
Building small AI tools and automations in Bengaluru. Notes on what ships, what fails, and what I'm learning along the way.