Amazon Best Seller Rank Prediction Using Multimodal Machine Learning

An Amazon Best Seller Rank (BSR) prediction system that uses a two-stage multimodal ML pipeline — first classifying whether a product will rank, then predicting the actual BSR value — powered by per-category XGBoost models trained on ~80 features from CNN/CLIP image embeddings, image quality metrics, NLP text analysis, and product metadata, all served through a FastAPI + Next.js full-stack web app where users can submit product details and images to get predictions.