What happens when you ground Google's MedGemma in 2,800+ pages of trusted clinical protocols? It cites its sources. It stays in scope. And it knows when to escalate. Running locally — no cloud, no API, no data leaving the machine.
NOT MEDICAL ADVICE: This is a research and education demonstration. MedGemma is not validated for clinical use. Do not use this for medical advice, diagnosis, or treatment decisions. Always consult qualified healthcare professionals.
This system retrieves content from the Community Health Aide Manual and generates AI-assisted interpretations. Responses may be inaccurate, incomplete, or misinterpret the source protocols. Always verify against the original manual and clinical judgment.
This is an independent research demonstration by Chief Health AI (PBC). It is not endorsed by, affiliated with, or approved by Google, the Indian Health Service, the Alaska Native Tribal Health Consortium (ANTHC), or any tribal health organization.
No conversation data is stored, logged, or transmitted to any third party. All processing occurs locally.
RAG (Retrieval-Augmented Generation) grounds an AI model in trusted source material. Instead of relying solely on what the model learned during training, RAG retrieves relevant documents at query time and uses them as context. The model answers based on evidence it can cite — not just patterns it memorized.
YOUR QUESTION
You ask a clinical question
VECTOR SEARCH
Finds relevant sections from 2,800+ pages of clinical protocols
MEDGEMMA
Reads the protocols and generates a grounded response
RESPONSE + CITATIONS
Answer with CHAM citations and escalation flags
The Community Health Aide Program (CHAP) was developed in the 1960s in response to health crises in remote Alaska Native communities. Formally recognized in 1968 and permanently authorized by Congress in 1992, approximately 550 Community Health Aides serve 170+ remote villages across Alaska.
CHA/Ps follow the Community Health Aide Manual (CHAM) for every patient encounter — it defines their scope of practice. The CHAM contains structured decision trees for every clinical protocol: symptom assessment, history questions, physical examination steps, plans, and explicit escalation criteria.
The Indian Health Service pioneered telehealth using NASA technology to serve remote tribal communities decades before the public internet. Congress authorized expansion of the CHAP to the lower 48 states, and the National CHAP Board was established in 2024 to oversee that expansion.
This project is inspired by the CHAP's innovation — not affiliated with it. It is not endorsed by, affiliated with, or approved by the Indian Health Service, the Alaska Native Tribal Health Consortium (ANTHC), or any tribal health organization.
Today, Local Health AI is grounded in the Community Health Aide Manual. The architecture is protocol-agnostic — the same RAG pipeline could retrieve oncology guidelines, pediatric protocols, emergency decision trees, or your organization's clinical pathways. Swap the knowledge source, keep the infrastructure. Local. Private. Evidence-grounded. Cited.
Interested in exploring RAG-augmented local AI with your own clinical protocols?
Explore Programs →NOT MEDICAL ADVICE: This is a research and education demonstration. MedGemma is not validated for clinical use. Do not use this for medical advice, diagnosis, or treatment decisions. Always consult qualified healthcare professionals.
This system retrieves content from the Community Health Aide Manual and generates AI-assisted interpretations. Responses may be inaccurate, incomplete, or misinterpret the source protocols. Always verify against the original manual and clinical judgment.
This is an independent research demonstration by Chief Health AI (PBC). It is not endorsed by, affiliated with, or approved by Google, the Indian Health Service, the Alaska Native Tribal Health Consortium (ANTHC), or any tribal health organization.
By using this demo, you agree to the Health AI Developer Foundations Terms of Use. View Terms
No conversation data is stored, logged, or transmitted to any third party. All processing occurs locally.
Built on MedGemma, an open model by Google. RAG pipeline powered by ChromaDB and nomic-embed-text.