Embedded Files
WHAT IS PROJECT RAIDEN-MED?
RAIDEN-Med is the evolution of project R.A.I.D.E.N., an agentic AI system to make access to specialist medical expertise more efficient, while ensuring results are aligned thanks to the feedback of a human medical doctor. The main differences are:
an update from Google Gemini 2.5 flash and flash-lite to Google Gemini 3.0 Flash Preview as LLM to manage the agentic workflow
the use of Google MedGemma instead of Google Gemini as model to represent specialists. MedGemma is a multimodal model specialized in medical reasoning.
RAIDEN-Med was submitted for the hackathlon of the MedGemma Impact Challenge on Kaggle.
If you want to know more about the previous version, project R.A.I.D.E.N., you can have a look here!
HOW WAS RAIDEN-MED DESIGNED?
First, the logo. It was made again with Google Nano Banana!
Concerning the main architecture, it is based on Google Gemini and the Google ADK for agentic AI. This time, a single specialist agent is created, and specialist medical doctors are tools of this agent. Such tools are powered by MedGemma. In this version of RAIDEN-Med, two specialists were created: a radiologist and a cardiologist.
The specialist agent receives the query with symptoms and provide a diagnosis. After that, there is a loop agent where a human doctor provide feedback on the diagnosis. If the human doctor approves it, a final report in both PDF and mp3 format is created.
A picture summarizing the architecture is below, together with a video showing how to use it.
Visual representation for the architecture of RAIDEN-Med.
Page updated
Google Sites
Report abuse