Fashion Recommendation System
Key Results
- 92% average precision in outfit detection
- Led cross-functional team of 3 engineers
- Increased customer click-rate and cross-sell
- Deployed across web and mobile platforms
The Problem
A major fashion retail platform needed to recommend complete outfits to customers based on individual items they were browsing. Existing rule-based recommendations had low engagement and failed to capture visual style compatibility.
The Approach
Led a team of 3 engineers to develop a computer vision pipeline using RetinaNet for garment detection and classification. The system identifies clothing items in product images, extracts visual features, and matches complementary pieces to generate outfit recommendations. OpenCV was used for image preprocessing and augmentation, with the model served via FastAPI in Docker containers.
The Results
The system achieved 92% average precision in outfit detection, significantly increasing customer click-rates and cross-sell conversions. The solution was deployed across both web and mobile platforms, processing thousands of recommendations daily.