Image-Based Recommendation System

I need a complete AI-driven recommendation system that learns directly from image data and returns high-quality, relevant suggestions in real time. The core of the job is to craft a model—preferably using Python, TensorFlow / PyTorch, and modern computer-vision techniques such as CNNs or transformers—that can extract visual features, compare them efficiently, and output ranked recommendations through a clean API. Here is what counts as done: • A reproducible training pipeline that ingests my image dataset, performs preprocessing and augmentation, then trains, validates, and saves the best model. • An inference service (REST or gRPC) that accepts an image or image ID and responds with a ranked list of recommended items plus similarity scores. • Brief but precise documentation covering setup, endpoints, and how to retrain or fine-tune with new data. • A short test script demonstrating end-to-end results on sample images. Clean, well-commented code in a Git repo and a quick hand-off call will wrap things up. If you have experience deploying vision models at scale and can explain trade-offs between accuracy, latency, and compute cost, that’s a strong plus.