DocNexus.ai · case study
Query-intent landscape
What do people actually ask a clinical search engine? Three methods, one vocabulary.
Three months of real search queries, triangulated three ways: local embeddings with UMAP and HDBSCAN clustering, a manual intent projection, and a transparent rule classifier.
Methods and relative results only. Client figures, internal names and scale are not published.
Result
Four capabilities explain ~87% of unique queries: entity resolution (43.0%), cohort discovery (35.1%), analytics (5.0%) and key-opinion-leader search (3.5%). The ambiguous remainder (11.2%) is reported, not hidden.
The useful finding was structural: the topic of a query and its intent are separate axes, and routing has to be built on intent.