Agentic Processing of Unstructured Health Data

Zoom information will be sent to e-HAIL members.

Healthcare generates vast amounts of unstructured and semi-structured text—diagnosis records, discharge summaries, and pathology reports, among others. Today, much of this information is still interpreted manually by clinical experts such as nurses and physicians, making the process slow, expensive, and difficult to scale. Recent advances in generative AI—especially agent-based systems—create an opportunity to automate portions of this workflow and meaningfully boost productivity.

Yet simply sending raw clinical documents to a model is rarely sufficient: context-length constraints, hallucinations, and limited domain grounding can lead to incomplete or unreliable outputs. To address these challenges, a growing body of agentic frameworks has emerged, combining techniques such as retrieval, decomposition, verification, and structured extraction to improve robustness and traceability.

In this talk, we use clinical trial matching as a running example to demonstrate how these frameworks can be applied in medical settings, and we evaluate their strengths and limitations in practice. We conclude by outlining our roadmap for a holistic agent-based framework for scalable, trustworthy unstructured health data processing.