Project · 2025
Fashion GenAI
Photo-in, outfit recommendations out
- Role
- Built the ML pipeline and API; React UI for upload and results.
- Context
- Fashion recommender that uses facial features and detected garments from an uploaded photo to suggest clothing.
- Timeline
- 2025
YOLO classifiers (type, color, pattern)
constraints & discovery
Inference had to run through a FastAPI backend; the browser only handles upload and presentation.
problems → solutions
problem 01 / 02
Multiple visual attributes needed consistent labels across uploads.
→ solution
Separate YOLO models per attribute with a single orchestration endpoint.
problem 02 / 02
Model output had to feel immediate in the UI.
→ solution
Async FastAPI handlers with clear loading and error states on the React side.
architecture decisions
- 01FastAPI service hosting Ultralytics YOLO checkpoints.
- 02React client for upload, preview, and recommendation display.
- 03Attribute detection decoupled from ranking logic so models can be swapped independently.
learnings
Splitting classifiers by attribute beat one monolithic detector for iteration speed and debuggability.
Next project
Cut invoice creation time by ~80%