Grounded Knowledge Assistant
A documentation assistant that answers only from what it can point to.
A retrieval-based assistant that answers questions grounded in a specific body of documentation, rather than from open-ended generation. It finds the relevant passages first, then answers from them — trustworthy enough to rely on for reference.
The problem
General chatbots invent answers. For reference use, a confident-but-wrong reply is worse than none. People need answers they can trust and trace back to a source.
The approach
Retrieval-augmented generation: index a corpus, retrieve the most relevant passages for each question, and constrain the model to answer from those. I prioritized grounding and a clean "I don’t know" over fluent-sounding filler.
The outcome
Grounded, checkable answers across a large document set, and less time spent hunting through the material by hand.