Core Team
Jakob Flunger
Agentic AI in Insurance
Master StudentAbout
Jakob builds agentic AI systems at the intersection of applied machine learning and real-world corporate deployment. Since the emergence of modern LLMs, he has continuously developed his own data and reasoning pipelines, progressively moving from static workflows toward fully agentic architectures. Earlier internships at Swiss Re and Mpreis grounded him in production software engineering. His technical foundation includes a BSc in Computer Science from ETH Zürich and master's-level coursework in Machine Intelligence, before pivoting to an MSc in Management, Technology and Economics at ETH. Beyond engineering, he has broadened his experience through a Customer Success Management role at Lyfegen, a pharma-adjacent Swiss startup. At ASL, Jakob develops an agentic AI sales assistant built as a joint venture between Zurich Insurance and Banco Sabadell to support bancassurance advisors. The system combines a tree-structured RAG knowledge base, a customer-profiling pipeline that turns CRM and transaction data into structured personas, and a propensity-scoring engine on section-level profile embeddings. His work sits between agent design, retrieval architectures, and the realities of deploying AI in a regulated corporate environment, where auditability and trust matter as much as raw model capability.
Research Areas
Connect
Project
Development and Testing of a Compliant Agentic Assistant for Life Insurance Distribution
This project develops an agentic AI copilot for insurance advisors in a retail-bank channel. The system helps advisors prepare for sales calls by generating structured briefings from CRM data and transaction histories, provides grounded, source-cited answers from product documentation during calls, and analyzes completed calls to improve targeting strategies and identify knowledge base gaps. A propensity-scoring engine ranks customers by product-specific conversion likelihood, enabling advisors to prioritize outreach across large uncontacted portfolios. From a scientific perspective, the project touches on data-structure-agnostic propensity modeling — building conversion predictions across heterogeneous inputs without heavy feature engineering — and on the dynamics of self-improving systems, where call-level feedback loops progressively refine both targeting and the underlying knowledge base. Developed with Zurich Insurance and Banco Sabadell, the architecture generalizes to any advisor-mediated sales channel where dense product knowledge, customer heterogeneity, and human-in-the-loop collaboration converge.
Other team members
Students
Interested in collaborating?
We are always looking for talented students, researchers and industry partners.
