At MindVista, I train image-segmentation models with YOLO to detect vehicle damage and evaluate DINOv2-based classifiers to measure how reliably they identify it. I focus on finding where models fail and reporting results clients can trust.
I also built and maintained Airflow ETL pipelines that onboard financial datasets onto a client’s data platform. For document intelligence, I developed a Claude-powered tool that reviews civil-engineering plans against regulatory checklists and produces reports.
My projects include BridgeRoman, an Urdu voice journal with an AI agent, and a fine-tuned vision-language model that transcribes scanned pages into Markdown and LaTeX. I take work from client scoping to delivery, including requirements, proposals, demos, and progress updates.
