Built an end-to-end image classification website using React (frontend), Flask (backend), and MongoDB (database). Used transfer learning (EfficientNet0) on ~14k images; achieved 96% accuracy. Dockerized and deployed on AWS LightSail.

Led a team of four as the lead developer to create a website for hosting tests, Q&As, and blogs targeting JEE and NEET aspirants. Designed the entire architecture from the ground up, implementing a React frontend, a Flask backend, and MongoDB for data storage.

Developed a pipeline to isolate a PAN card from its background and extract essential details. Utilized OpenCV for image processing and EasyOCR for accurate text retrieval.

Built a CNN model on Kaggle's Digit Recognizer competition to classify handwritten digits (MNIST dataset). Achieved a 99% accuracy rate, demonstrating strong model performance.

Shipped agentic chatbots using LangGraph with complex architecture for E-commerce/Real-estate/Govt. sectors cutting down cost on customer service and improving product discovery and sales.

Built RAG based chatbots with guardrails; reduced hallucinations by context engineering. Designed ingestion pipelines to convert large heterogenous data in excel sheet formats/pdfs/docx into Machine ingestible JSON/Text.