Goal

Build a prototype that allows users to quickly build a custom monitoring dashboard using analyses available via the data-api. Users should be able to describe the dashboard set up they need in order to monitor their portfolio (defining areas of interest, specifying data to analyse, and insights). The solution should enable insights across the portfolio as well as for individual areas.

The LLM will generate sql based on user prompts or map state, run analyses against precomputed tables or on-the-fly endpoints. Additionally, users will be able to interact with the LLM to tune the sql as necessary, summarise insight trends, explain patterns in the data or add context from descriptive datasets (PAs, KBAs, Project info)

Phase 1: Experimentation

• Design wireframes for dashboard construction, demonstrating key user interactions
• Experiment to test table selection and sql generation against precomputed GFW dataset tables based on user prompts
• Experiment to test chart selection based on resulting data
• Experiment to test insight interpretation using fixed sql queries (e.g. can the LLM summarise as good as GFW dynamic sentences?)
• Experiment to test if LLM can supplement query results with rich text data about areas of interest (e.g. execute batch queries against KBAs using data-api raster analysis, and incorporate this into interpretation)

Phase 2: Implementation

• FE implementation