I started by using the existing Smithsonian Open Access search tool the way a first-time visitor would — no species names, no taxonomic knowledge, just curiosity. The friction was immediate:

Large, Unfamiliar Datasets:
Search requires knowing scientific terms (genus, species epithet) that casual visitors don't have.

Fragmented Records: Records are shown one at a time, with no way to compare across the collection. Relationships that make orchids interesting to a lay audience — why a flower smells the way it does, when it blooms, who pollinates it — only become visible when you look across hundreds of records at once, which the interface doesn't support.


This shaped my core design question: 
how might we let a non-expert discover patterns in a scientific dataset without requiring them to know what to search for?