Mr001learner/VisMatch

🛍️ VisMatch - AI-Powered Visual Product Search

VisMatch implements a smart visual product search using a lightweight deep learning approach. The system uses MobileNetV2 to extract features and turns each product image into a 1280-dimensional vector. These feature embeddings get indexed and compared with FAISS (Facebook AI Similarity Search), allowing for visual similarity queries under 500 ms with over 75% accuracy.

🛠️ Tech Stack

The backend is built with FastAPI. It includes asynchronous routes, strong error handling, and logging for dependable production use. The machine learning pipeline combines PyTorch, MobileNetV2, and FAISS IndexFlatIP for real-time vector similarity calculations. A lightweight SQLite database holds 77 products and their metadata, including brand, price, category, and rating.

The frontend is a modern JavaScript-based interface that supports drag-and-drop uploads, image URLs, and a design that works well on mobile devices. Users can filter search results by category and similarity threshold.

The whole solution is packaged with Docker. It includes health checks and monitoring, making it easy to deploy on any cloud or free hosting platform.

📊 Product Categories

• —👜 Bags - Handbags, backpacks, purses
• —📱 Phones - Smartphones and accessories
• —👟 Shoes - Sneakers, boots, formal shoes
• —⌚ Watches - Smart watches, luxury timepieces

🔗 Links

• —Source Code: GitHub Repository
• —Developer: Abishek - Full Stack AI Developer

📄 License

Apache 2.0 License - Feel free to use and modify!