Things I've
built and shipped.

End-to-end academic research automation — retrieves papers from arXiv, processes them through a RAG pipeline, and generates structured literature reviews on demand. Cuts hours of manual reading into a repeatable, queryable workflow.

Adaptive quiz generator that ingests text or PDFs and generates questions calibrated to Bloom's Taxonomy cognitive levels. Includes a real-time analytics dashboard for tracking learner performance over time.

A Streamlit app that lets users upload documents and chat with the content using LLMs. LangChain handles document retrieval and prompt chaining; Hugging Face transformer models power semantic search. Built for teams that need to query internal knowledge without manual searching.

Object detection model trained to identify dents and scratches on product surfaces for manufacturing QA. Collected and annotated image datasets, performed image preprocessing with OpenCV and Pillow, trained in Google Colab, and tuned hyperparameters — evaluated using mAP for real-world inspection accuracy.

Multi-modal ISL system that converts text, images, video, and speech into sign language animations. Sign recognition is built on a CNN model with MediaPipe hand tracking. Handled data preprocessing, animation sequencing, and evaluation metric design. Published at ICCCNT 2023.

CNN trained to classify handwritten mathematical symbols and digits. A Flask web app lets users draw equations on a canvas in real time and receive formatted LaTeX output — handling image segmentation, symbol classification, and dynamic LaTeX generation end to end.