Skip to main content
Cora is haau3’s conversational AI. It helps caregivers and patients understand their own health information in plain language, and it is built to stay within strict safety limits. Cora is haau3’s entry in the CMS Health Tech Ecosystem conversational-AI category: AI that supports health literacy without giving medical advice.

​
Safety first

Cora operates under fixed safety rules. In practice this means Cora:
  • Gives educational information only, never medical advice.
  • Does not diagnose, predict what will happen, or recommend treatment.
  • Does not judge whether a value is good, bad, normal, or dangerous beyond what the record itself already states.
  • Points people to their doctor, nurse, or pharmacist for anything about what their information means or what to do, and to 911 for anything that sounds like an emergency.
  • Says it is an AI assistant when asked.
Cora is grounded: it works from the information in the person’s own record and from vetted public-domain reference material, and it does not invent values, ranges, or dates.

​
Languages

Cora supports English and Spanish today, with a bilingual option. The design is built to extend to the other major US languages with more than one million speakers (Chinese, Tagalog, Vietnamese, Arabic, French, Korean, Portuguese, Hindi, Haitian Creole, and Russian) through model-assisted translation. Those are on the roadmap and are not yet available.

​
Models and compliance

Cora runs on Google Gemini models: gemini-2.5-flash for text and gemini-live-2.5-flash-native-audio for voice. It operates under a HIPAA Business Associate Agreement, with no consumer AI path, so a patient’s data stays within the compliant boundary.

​
Nothing is stored

Cora is ephemeral. Conversations are not saved. No transcript, audio, typed text, or model output is persisted or logged anywhere. A session exists only while it is open, and nothing about it remains afterward.