AI Facet Selection PoC #4609
Description
Background
Users often begin searches with unstructured, natural-language questions that encode multiple structured parameters (e.g., disease, datatype, consent, study design). Currently, they must translate these manually into facet selections.
By integrating an LLM-based intent parser into the Data Explorer, we can allow for intuitive natural-language queries while retaining the integrity and transparency of our existing filter system.
Goals
- Let users describe their intent in natural language (e.g., “whole genome data on T2D with GRU consent”).
- Automatically map free-text input to structured facet filters.
- Provide clear explanations of how query terms were matched to canonical metadata.
- Maintain faceted filters as the authoritative state of active search parameters.
- Preserve transparency and user trust through visual mapping cues (e.g., “interpreted from…” lines).