Wenqi Shi, PhD

Assistant Professor

Department of Health Data Science and Biostatistics

UT Southwestern Medical Center

Talk Title: Scaling Biomedical Intelligence in LLM Agents: Toward an Integrated Environment for a Self-Improving AI Co-Scientist

Abstract: Progress toward an AI co-scientist, an agent that formulates hypotheses, executes analyses, and reasons toward discovery, is constrained less by the reasoning capacity of large language models (LLMs) than by the infrastructure in which they operate. Biomedical data science research remains fragmented across specialized knowledge sources, heterogeneous databases, and non-interoperable tools, so that even capable models yield analyses that are difficult to reproduce and verify. We propose an integrated environment for agentic biomedical data science that unifies grounded evidence retrieval, executable analysis over real biomedical data, and verifiable feedback into a continuous self-improvement loop. By learning from measurable outcomes rather than unverified self-assessment, the agent can iteratively propose, test, critique, and refine its solutions. Such an environment can serve as both a workspace and a training ground for scalable, reproducible biomedical intelligence, advancing the next generation of AI co-scientists.

Host: Dr. Ting Wang