I Built a Local Clinical AI Agent from Scratch — Here’s How

Author(s): Marie Humbert-Droz, PhD

Originally published on Towards AI.

How I wired GPT-OSS with custom tools to make clinical data actually usable.

In my last experiment, I ran OpenAI’s new local model on my laptop and it extracted JSON from clinical notes surprisingly well. However, these raw extractions are not very actionable. We successfully got a lot of information out of a discharge summary, like the primary diagnosis, relevant lab results and follow-up appointments scheduled. But this information is not actionable yet: we need clinical reasoning, normalization, and interoperability.

This article details the development of three tools built to enhance the usability of clinical data extracted from AI models, focusing on transforming raw outputs into structured, EHR-ready information through processes like abnormal lab detection, follow-up gap checking, and diagnosis normalization with UMLS, ultimately demonstrating the effectiveness of local AI models in clinical reasoning.

Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI