Trendbook
Ongoing•Deep Learning
Trendbook is my largest solo-developed project. It’s a complex technological platform that combines a domain-specific CNN model deployed to production with a scalable, advanced backend for processing big data efficiently - all wrapped in a simple yet aesthetic mobile app. This is my proudest project to date.
Technologies Used
PythonPyTorchFastAPIAWSCNNVector DatabasePostgreSQLQdrant
Project Overview:
Trendbook is an AI-powered visual search platform for fashion. Users simply snap a photo, and our proprietary AI lens analyzes the image to detect clothing items and instantly finds matching or similar products from multiple retailers, complete with price comparisons.
Key Contributions & Highlights:
- Developed a custom CNN model for cross-domain fashion image retrieval, leveraging a feature pyramid strategy and parallel convolutional blocks to better capture clothing semantics in real-world photos and product images.
- Built a high-performance inference pipeline using FastAPI and PyTorch, hosted on autoscaling AWS GPU instances with load balancing to handle real-time global searches.
- Engineered a robust search backend integrating a vector database for similarity search and PostgreSQL for product data, with intelligent caching of recent searches to reduce GPU costs and deliver instant results.
- Designed and maintained a custom feed sync API, processing over 10 million SKUs daily from hundreds of supplier feeds, supporting product variants and multi-merchant listings.
- Delivered a seamless mobile app experience, enabling thousands of users to discover and purchase fashion items with just a photo.