GenAI Customer-Support Assessment
A customer-support assistant that routes each question deterministically — SQL for data, RAG for documents, hybrid for both — and cites where every answer came from.
A support question goes in; a complete answer comes out with its source attached. The assistant decides on its own whether the question needs the database, the document corpus, or both — no mode switch, no dropdown, no asking the user to know the difference.
Ask a data question, a policy question and a mixed one — each visibly takes a different path: SQL for data, RAG for documents, hybrid for both. Routing is deterministic, so the same kind of question takes the same route every time.
A knowledge-base hit comes back with the answer and its basis — the chunks it drew from, cited explicitly, so a support agent can check the source before forwarding the answer to a customer.
Ask something outside the assistant's scope and it says so explicitly instead of improvising a confident guess. In a support context a wrong answer costs more than no answer — the fallback is designed behavior, not an error state.
Ask a follow-up in the same conversation and the context carries — the assistant answers the new question in light of the last one, without being re-briefed on what was just discussed.
The Playwright checks drive the real app — the flows on this page are the flows the tests walk, and the suite ends green on camera.
Data questions hit the database directly; document questions stay grounded and cited; mixed questions get both. One assistant, the right tool each time.
Every response carries an explicit citation back to the rows or chunks it used — answers a support team can forward without re-checking the source themselves.
Out-of-scope questions get a designed fallback, not a confident guess.