About this assistant
An assistant that answers plain-language questions from 's own data, with numbers, tables, charts and maps.
What it is
The assistant runs on RDDSquare. Staff ask a question in their own words; the assistant works out which dataset can answer it and returns the answer, with a map, chart or CSV download when asked.
How it works
Each dataset the assistant may use has a reviewed description: its fields, what they mean, the coded values and how its records link to map features. The language model reads the question and turns it into a request against those descriptions. The numbers come from a database query the engine builds from the reviewed description; the model does not write free-form queries. A question that no reviewed description covers is declined rather than guessed.
Consistent answers
The same question asked the same way gets the same query and the same numbers. Every answer says which data and filters it used, so it can be checked.
Data sources
The assistant reads the data from 's database tables and views and its map services; it never changes them.
Safety & privacy
The assistant reads data through a read-only account and only the tables and views approved for it. Map layers are read with each user's own permissions. Questions and answers are logged in the site's own database for review.
Limits
- Answers are only as complete as the data behind them; when a field is often empty, the answer says so.
- The assistant informs decisions; it does not make them.
Feedback
Rate any answer with the thumbs up or down buttons and add a comment. Every comment is read.