DAE Search
Enterprise search redesigned around an inconvenient truth: finding the data is only half the job. Knowing whether to trust it is the rest.
The Problem
Analysts search "revenue," get 40 results, then spend 20 minutes figuring out which table is current, which is the team-of-record's, and which was deprecated three quarters ago but never cleaned up.
The job isn't returning results — it's returning the one result you can act on.
Image Failed to Load
This image is currently unavailable
What I Built
Semantic search over metadata, not keyword match — tables called `arr_monthly` surface for "revenue." Results cut from 40-to-narrow-down to 4-to-pick-from.
Data lineage lives on the result itself: where the data came from, when it last refreshed, what depends on it. One glance, not a click-through.
Image Failed to Load
This image is currently unavailable
The Decisions That Mattered
Permission state is a first-class signal: restricted results stay visible with a lock and a one-click access request. Hiding them entirely made people think the data didn't exist.
Permission-aware autocomplete was built, security flagged it — the suggestion box was leaking the existence of restricted datasets through pattern-matching.
Image Failed to Load
This image is currently unavailable
Image Failed to Load
This image is currently unavailable
What I Got Wrong First
Started by treating this as consumer search with an enterprise wrapper — clean ranked list, minimal chrome. Wrong audience.
Enterprise users want context, signals, and density. Analysts type fragments and abbreviations, not full questions.
Image Failed to Load
This image is currently unavailable
Outcome
The shift from keyword match to semantic search with lineage and permissions inline reframed the product from a search tool to a data discovery tool.
In enterprise contexts, trustworthiness of a result matters more than relevance. Most search UX optimizes for the second.