At eBay, I built high-precision multimodal models that match seller listings to catalog products for recommendations, boosting GMV by $20 million. I also improved catalog-matching precision by 2.5%, coverage by 4%, and recall by 2% across multiple categories.
I'm researching and fine-tuning vision LLMs for fine-grained image extraction in the Collectibles segment. I optimized and deployed LLMs to improve throughput by 30%, and prototyped a RAG-based buyer assistant for personalized recommendations and agentic assistance.
At Wells Fargo, I built a RAG-based LLM assistant for merchant workflows from prototype to production, reducing loan application time by up to 30%. I also engineered credit-card services processing more than 1 million transactions per month across 18,000+ merchants.
My research includes continual object detection for autonomous driving at Mila / Robotics and Embodied AI Lab, University of Montreal, where my undergraduate thesis work cut catastrophic forgetting in object detection by 73%. I also proposed the AdQuesta VQA framework, which doubled QA accuracy compared with previous state-of-the-art approaches.
