Jasmin Patel

Building a production-grade Invoice classification system.

The core ML pipeline uses TFIDF vectorization + Adaptive Random Forest Classifier (River), an online learning approach where the model updates incrementally with each new invoice, no full retraining required.

Stack: FastAPI, River, MySQL, Docker Compose, Ollama, AWS EC2 , Python

Also integrating Qwen2.5 VL, a multimodal LLM, for structured data extraction directly from invoice images.