All projects
October 2024
StyleNova AI — Vision-Language Adaptive Fashion Recommender
Computer VisionRecommendation SystemsDeep LearningE-commerce
- Fashion recommendation engine that learns your taste fast — combining what an outfit looks like, what you've liked before, and what similar shoppers bought — to suggest items you'll actually want.
- Hybrid scorer fusing CLIP ViT-B/32 (512-dim), TF-IDF content signals, and user-based kNN collaborative filtering achieves Precision@10 of 0.73.
- Online preference learning via EMA (λ=0.3) with ε-greedy exploration improves recommendation relevance 32% after just 5 user interactions.
- Pre-computed CLIP embeddings cached as SQLite BLOBs deliver sub-200ms p95 API response for 10K+ concurrent users — zero cold-start latency.
PythonPyTorchCLIPFastAPIscikit-learnNext.jsTypeScriptPrismaZustandFramer Motion