Building Systems. Refining Architecture. Deploying AI.

A Retrieval-Augmented Generation (RAG) web app that lets you upload a document (.txt or .pdf) and ask questions about it in a ChatGPT-style chat interface. Chunks and embeds the document with HuggingFace embeddings, stores them in Chroma, and generates grounded answers using Azure OpenAI — all served through a Flask backend.

Built an end-to-end medical image analysis prototype for binary classification (normal vs abnormal) on grayscale X-ray/MRI scans using a custom 4-block CNN architecture in PyTorch. Designed the training pipeline with augmentation, class balancing, and early stopping, achieving 95% validation accuracy.