5 réponses
Hi Raju,
Recently we have built a Tableau Dashboard Extension that adds a chat panel right inside the dashboard — so instead of eyeballing charts, users can just ask questions in plain English ("what drove the drop in Q2 expenses?", "compare loan fees to expense spend") and get grounded, numeric answers.
How it works:
- One lightweight local/hosted backend serves multiple dashboards at once — the model automatically figures out which dataset(s) a question is about, and can even blend two datasets in one answer.
- Every answer is computed by the LLM writing read-only SQL against a real snapshot database (using Tableau's own Hyper engine) — not hallucinated numbers, not a live query against the workbook. This keeps answers fast, accurate, and safe even on 10M+ row extracts.
- Each dataset ships with auto-generated reference docs (a description + data glossary) so the model understands field meanings, grain, and caveats before answering.
- An admin flow can onboard a brand-new Tableau Cloud workbook automatically: pull it via the REST API, extract the embedded data, and have the LLM draft the descriptive docs itself.
- Row-level-security workbooks are auto-detected and blocked from onboarding unless explicitly overridden, so a shared snapshot never leaks restricted rows.
Please upvote this response if this will be helpful to you.
Regards
Rajneesh
