Topic hub
RAG
RAG only helps when retrieval finds the right evidence, fits it into context, and proves the answer follows from the source. The useful work is retrieval quality, chunk audits, and citation discipline.
Start here
- →Evaluate retrieval separately from generation.
- →Inspect chunks before blaming the model.
- →Validate citations against the underlying source.
- →Treat missing evidence as a real product state.
Related writing
-
Essay · ai · rag · evals
How Frigade Slashed Latency & Boosted User Helpfulness
-
Essay · ai · rag
The Enduring Value of Retrieval in the Era of Expanding Context Windows
TILs
-
TIL · ai · rag · evals
A bad RAG answer does not tell you whether retrieval failed, generation failed, or the product asked an impossible question. Split the blame before fixing anything.
-
TIL · ai · rag · evals
Before blaming the model, inspect the chunks. Duplicate, empty, bloated, or low-signal chunks can wreck retrieval quietly.
-
TIL · ai · rag · evals
A citation is not proof just because the model printed a source name. Verify that the source exists and actually supports the claim.