DHRUVKAUL

I owned the end-to-end build of a full-stack AI sponsorship analytics platform for a leading global sports brand (under NDA), measuring how its products show up across athlete social content spanning 30+ athletes in 7 sports. Built with Python, GCP, Gemini Vision, and TypeScript, the computer vision pipeline matches exact products against a 6,000+ product catalog covering 40+ footwear and apparel lines, with an image segmentation stage driving 73% of correct identifications. I also engineered a corrective agentic AI layer over MCP that catches missed and inaccurate detections through image analysis, web search, and product verification, improving detection accuracy by 19%, and shipped the client-facing dashboard that presents the insights.

I built Java backend services that automated loan approval review, increasing application throughput and contributing to roughly 20% more approvals. I also developed Python pipelines on AWS that predict credit card application likelihood for 1.2M+ customers, enabling more targeted mailing campaigns.

I worked with researchers and historians on an AR app that surfaces 3.6M+ digitized cultural artifacts from national archives. I designed the computer vision model that reads image-embedded scale bars to estimate real-world artifact dimensions, improving true-to-size AR display accuracy by 53%.

I led a team of 5 building a GenAI-powered Q&A chatbot for an anime and manga Web3 platform backed by Japan's $1B "DigiDen" initiative, lifting user engagement by 35%.