Tryora

Problem

Online shoppers can't tell how a garment will actually look on their body before buying, and generating photorealistic 3D avatars or AI try-on scenes is too GPU-intensive to run synchronously behind a normal web request.

Solution

Built a distributed system that turns user photos into a 3D avatar, matches natural-language queries to real dresses via an LLM-driven scraping pipeline, and composites the avatar into AI-generated try-on scenes, all coordinated through an Express API gateway that dispatches GPU-bound jobs to a FastAPI/Celery worker over a Redis queue, with offline access to saved wardrobes via a PWA service worker and IndexedDB.