Problem
What made this hard
Traditional search depended on exact metadata. People needed to describe a scene, object, or moment in ordinary language and still understand why each result appeared.
All case studies04 / 11
AI discovery · Enterprise storage · 2023
IRIS AI Discovery
Find the right media through a conversation and inspect the evidence.
RoleSenior UX Engineer IIStatusShippedContextDataCore · IRISEvidenceNatural language discovery · Faceted refinement · Timestamped recognition evidence
Outcome
What changed
A conversational search model that moves from an open prompt to precise filters, relevant results, and timestamped recognition evidence.
The thesis
An AI discovery experience for searching large media collections with natural language, structured facets, and explainable result evidence.
Case-study proof
The decision record
Enough context to judge the choice, inspect the work, and separate evidence from ambition.
Why this, not that
I combined natural-language discovery with familiar facets and timestamped reasons instead of choosing either rigid metadata search or an opaque conversational answer. People could start with memory, then narrow and verify the match.
Why this case is specific
IRIS searched large media collections where a result could depend on object recognition at an exact moment—not just a filename or a document keyword.
What you can inspect
Shipped product screenshotsThe evidence sequence shows a natural-language request, result groups, follow-up refinement, and the recognition evidence tied to timestamps.
Decision to outcome
The shipped design connected conversational intent to inspectable search evidence. It supports a clear trace from query to matched moment; the page does not claim unmeasured improvements in search speed or accuracy.
Interface evidence
01 / Intent
Let people search the way they remember
The prompt begins with a scene or object in natural language. Suggested paths help people form a useful query without forcing them to learn a search language.
02 / Precision
Conversation opens into control
Facets, result groups, and clear query context let a broad request become precise without starting again.
- Natural language prompt
- Suggested discovery paths
- Faceted result grid
- Persistent query context
03 / Trust
Every result needs a reason
A reason panel connects the request to matched metadata and object recognition. Timestamped evidence lets people verify the exact moment that produced the match.
04 / System
Reuse familiar search behaviour
The experience combines new AI capability with familiar search patterns, helping people explore without surrendering control.
Supporting evidence
Bring me the difficult part.
Complex workflows, AI trust, and enterprise systems.
Discuss a product challenge