Jisheng Jiang

At The Paddock, I expanded the product from cars to motorcycles while preserving existing automotive features. I also reworked social graph pagination and exact-count queries, reducing follower and following payloads by 99.7%. At University of Illinois Urbana-Champaign, I’m developing an LLM- and knowledge-graph-driven personalized learning system, and at AI Alignment @ Illinois I built a multi-agent debate framework that improved GSM8K mathematical reasoning accuracy by more than 40%. Previously, I developed backend systems at Tripalink and stateful LLM infrastructure at VideoTutor, including retrieval and model routing for a product that reached 30K+ users.