Pragmatic RAG: moving data safely between services
Combining services with retrieval-augmented generation—without leaking data or hallucinating answers.
RAG shines when you need to combine knowledge from multiple sources into a coherent answer. The risk is pulling in data the user shouldn't see, or generating confident nonsense.
Pragmatic RAG starts with strict retrieval boundaries: scope embeddings per tenant, filter at query time, and never trust the model to self-censor.
I like building agents that orchestrate APIs and RAG pipelines where data boundaries are enforced at the retrieval layer—not hoped for in the prompt.