Care Compass Pairs Gemma 4 With Policy Evidence

Why this is here: The system creates a forensic decision record containing verified policy artifacts, selected rule IDs, and a hash of the model output for every interaction.

Copyleftdev built Care Compass, a local healthcare navigation demo that uses Gemma 4 and Aion to verify signed policy evidence before answering questions. The project aims to address the problem of untrustworthy AI responses by creating an inspectable record of every decision. Care Compass functions as a console for scenarios like discharge follow-up and low-cost clinic searches, but limits Gemma 4 to language tasks within defined boundaries.

The system uses a deterministic gate to allow, block, or escalate requests based on pre-defined policies covering safety signals, clinical scope, and privacy. When a request is approved, Gemma 4 generates helpful information.

Unsafe requests bypass the model entirely. The developer provides a Docker and Ollama demo, with options to run with smaller or larger Gemma 4 model profiles.

The project’s code is available on GitHub, emphasizing inspectability and a focus on creating evidence of decision-making rather than relying on model explanations. Further development would require clinical review, accessibility testing, and stronger resource verification.