I translated evolving AI search, multimodal discovery, creator workflows, and personalization capabilities into one coherent consumer experience for a platform serving 300M+ monthly active users.
Users should not have to understand which model, surface, or system produced an answer.
AI search, visual understanding, generation, publishing, and personalization had different inputs and failure modes. The experience needed a common interaction grammar without flattening what each capability did best.
Text, screenshots, photos, products, personal taste, and creator intent all enter the system differently.
Recommendations and generated outputs need source transparency, confidence cues, and inspectable reasoning.
Discovery only becomes valuable when intent, saved context, and edits persist into later actions.
I modeled the ecosystem as a stateful loop, not a collection of AI screens. Each step produces structured context that the next surface can reuse.
The prototype connected entry points, AI actions, state transitions, and downstream destinations so teams could judge the ecosystem as one journey.
Fashion Memory makes what the system knows visible, editable, and useful across future sessions instead of hiding personalization inside a black box.
The interaction sequence makes AI interpretation inspectable before the user commits to a generated look or publishing action.
Low-confidence visual attributes are marked for review instead of being presented as facts.
User corrections update the local preference model and explain what will change next time.
People can revise one constraint, preserve the rest, compare alternatives, and return to an earlier state.
I partnered with product, engineering, AI, and recommendation teams to define how model states should appear, what control users receive, and how the system recovers when confidence is low.