Responsible AI
Hekavor uses AI to help physicians take a thorough history before the patient's visit — never to replace their clinical judgment. This page explains how we build and operate that AI responsibly, and the limits we place on it. Our approach aligns with widely recognized frameworks for responsible health AI, including the Coalition for Health AI (CHAI) and the Joint Commission's guidance on the Responsible Use of AI in Healthcare.
Physician-authored, not AI-improvised
Hekavor's assistant — Mari, parts of which are AI-enabled — does not autonomously generate clinical interview questions or make clinical diagnoses. Its interview pathways consist of structured questions that are written and reviewed with physician input. AI is used in a limited, supporting role to facilitate information collection and workflow efficiency. For example, AI may be used to interpret a patient's free-text response and map it to the structured question being asked, helping to facilitate information collection. AI is not used to independently determine what to ask or reach a clinical diagnosis on its own.
Patients presenting with the same chief complaint are generally guided through a standardized pathway to promote consistency in information gathering while supporting clinician oversight and judgment.
A physician is always in control
Mari produces decision support, not decisions. It does not provide a diagnosis or medical advice. Every output — the history, the working differential, and any "do-not-miss" flags — is presented to the physician to review and edit before it is used or entered into the record. The treating clinician makes all clinical decisions.
Transparency for patients
Patients are told they are interacting with AI. Mari introduces itself as an AI-powered interview assistant working on behalf of the physician's practice before the interview begins — it does not pose as a human clinician. Completing a Mari interview is voluntary: a patient who would rather not use it can give their history directly to their physician instead, with no effect on their care. Patients are also provided with clear and accessible notices regarding our use of AI and any applicable terms, privacy notices, and consent requirements. For details, see our Terms of Service and Privacy Policy.
How we build and validate Mari
Mari's clinical pathways are developed and reviewed by physicians. Prior to deployment, pathways are subject to testing and validation informed by physician-developed clinical scenarios. We validate a pathway's behavior before it is deployed in a clinical environment, and we monitor how Mari performs once it is in use, so that changes are caught, reviewed, and corrected. Mari augments a physician's history-taking; it is not a diagnostic device and does not make autonomous clinical decisions.
Fairness and reducing bias
Because Mari is designed around physician-authored pathways, patients with similar chief complaints are typically guided through the same structured and clinically informed interview process. This approach is intended to promote consistency and support equitable information collection across patient interactions. We design and review these pathways to serve the diverse patients a practice sees, and we monitor for differences in how Mari performs across patient populations so that we can identify and address inequities. AI can reflect biases present in data and in medicine; we treat reducing that risk as an ongoing responsibility, not a one-time check.
We minimize what AI ever sees
We do not send patient contact information to any large language model. A patient's name, email address, phone number, and home address are never included in prompts to an AI model. The AI works with the clinical content of the interview — not the identity of the person.
We do not train AI on your data
We do not use patient data, interview conversations, or any information you or your patients submit to train AI models, and we do not send it to model providers to be used for their training. We may use limited, de-identified information to provide, improve, and maintain the services we provide you. Your data is used to deliver the service to you — not to build models.
Accuracy and limitations
AI systems can make mistakes. Mari's differential and flagged findings are suggestions intended to support a physician's review — not autonomous conclusions, and not a substitute for clinical examination and judgment. Physicians should evaluate every output in the context of the patient in front of them.
Security and privacy
Mari is built for healthcare. Patient information is encrypted in transit and at rest and protected with permission-based access controls. We sign a Business Associate Agreement with every practice before any patient data is exchanged, and we handle protected health information in line with HIPAA. For details, see our Security overview, Business Associate Agreement, and Privacy Policy.
Oversight, monitoring, and accountability
Responsible AI is not a one-time checklist. A physician-led team owns Mari's clinical pathways and the practices described on this page — reviewing and maintaining them, and monitoring Mari's behavior and outcomes so that it stays consistent, safe, and within the bounds described here. When we identify an issue, we review it and correct it.
Education and training
We work collaboratively with clinicians to support the safe, effective, and responsible use of our AI solutions through ongoing education, training, and workflow oversight. Our engagement is continuous rather than a one-time deployment. We maintain current knowledge of evolving clinical practices, AI technologies, and regulatory expectations by engaging in professional development, industry forums, conferences, and ongoing engagement with the healthcare community.
Questions or concerns
If you have a question about how Hekavor uses AI — or want to report a concern or a possible AI safety issue — email us at contact@hekavor.com. We review every report.