AI4RD

As genome sequencing goes mainstream in healthcare, the wrong test is being ordered more and more often, which is an expensive, time-consuming mistake with significant negative impacts on the patient.

To address this need, we developed Test Suggest, an AI agent that guides the clinician through the test ordering process. This solution extends our existing health software already in use with genetics clinics in multiple countries.

The app is written in Python using streamlit and gemini-2.5-pro, with the test directory provided as context along with a system prompt. This was compared to Vector AI Search and a custom Gemini GEM. We tested it on a variety of manually generated prompts, with a system prompt and code developed with the support of cursor.

The next step of the development is to provide the model with a tool that allows the user to order the preferred genetic test. This then feeds into a genomic laboratory user interface in which human feedback is provided and the correct test provided. This is provided as additional context to the suggestion model to improve future suggestions. Future improvements include increasing the speed of the application using pre-caching of the context files, and reimplementing using ADK.

This application is a prototype implementation of a clinical tool with significant benefit to health care providers, by helping make genetic testing and personalized medicine more accessible to less specialized providers. This is the future of healthcare.