Core Team
Connor Larson
Agentic AI in Insurance
Master StudentAbout
Connor Larson is a Master’s student at ETH Zurich working at the intersection of mathematics, finance, and applied AI. In the ASL, he focuses on agentic AI systems for insurance, exploring how large-model-based workflows can automate complex, document-heavy processes in collaboration with Zurich Insurance. His work sits at a particularly interesting boundary between frontier AI research and highly consequential enterprise operations, where reliability, interpretability, and workflow design matter as much as raw model performance. Before joining ETH, Connor studied Operations Research and Financial Engineering at Princeton, building a foundation in quantitative decision-making, optimization, and real-world problem solving. He also brings experience from management consulting, private equity, and data science, which gives him an unusually sharp perspective on how technical systems create value inside large organizations. At ASL, he combines analytical rigor with practical market intuition to work on AI systems that are not only technically ambitious, but also deployable in industries where efficiency, accuracy, and trust are essential.
Research Areas
Project
Agentic AI for Commercial Insurance Workflows
Commercial property and casualty insurance remains one of the largest and most operationally complex markets in the world, yet many core processes still depend on fragmented communication across phone calls, email threads, spreadsheets, and PDF documents. This project investigates how agentic AI systems can orchestrate these workflows more intelligently by combining language models with structured reasoning, document understanding, and task-level decision making. Rather than treating insurance automation as a single prediction problem, the work approaches it as an end-to-end operational system in which AI agents can assist with intake, triage, information extraction, case handling, and coordination across multiple stakeholders. Scientifically, the project is interesting because it examines how modern AI systems behave in environments that are unstructured, high-stakes, and operationally constrained. Insurance workflows require models to interpret heterogeneous documents, maintain context across long task horizons, reason under uncertainty, and interact with human users in ways that are auditable and reliable. This creates a rich research setting for studying agent design, tool use, workflow planning, and human-AI collaboration in enterprise environments. The project also raises broader questions around robustness, controllability, and evaluation: not simply whether an agent can complete an isolated task, but whether it can support a multi-step business process with consistent quality. In practice, this work addresses a large and urgent market need. Commercial insurance is a massive global sector where inefficiencies in back-office and customer-facing operations translate directly into cost, delay, and avoidable risk. Systems that can reduce manual handling, accelerate turnaround times, and improve consistency have clear value for carriers, brokers, and enterprise clients alike. By focusing on a workflow category that is both economically significant and technically underserved, the project points toward AI infrastructure that could become foundational for next-generation insurance operations and adjacent document-heavy service industries.
Other team members
Students
Interested in collaborating?
We are always looking for talented students, researchers and industry partners.
