TLDR: Agentic-AI Healthcare is a research prototype that integrates intelligent agents, robust privacy features, and multilingual support (English, French, Arabic) for digital health interactions. It uses the Model Context Protocol (MCP) to orchestrate agents for symptom checking, medication suggestions, and appointment scheduling. A dedicated Privacy & Compliance Layer ensures data protection through AES-GCM encryption, role-based access control (RBAC), and tamper-evident audit logging, adhering to HIPAA, PIPEDA, and PHIPA standards. The system aims to provide explainable AI outputs and address limitations in existing conversational healthcare tools regarding privacy, transparency, and language accessibility.
A new research prototype, Agentic-AI Healthcare, introduces a groundbreaking approach to digital health by combining intelligent agents, robust privacy features, and multilingual support. Developed as a single-investigator project by Mohammed A. Shehab from Concordia University, this system aims to address critical limitations in existing conversational healthcare platforms, such as privacy concerns, lack of explainability, and limited language accessibility.
At its core, Agentic-AI Healthcare leverages the Model Context Protocol (MCP) to orchestrate multiple intelligent agents. This protocol allows for the seamless coordination of different AI modules, each responsible for a specific aspect of patient interaction. The current prototype features three key agents:
Symptom Checker Agent
This agent is designed to analyze natural language descriptions of symptoms provided by patients. It extracts structured information and generates a preliminary diagnostic hypothesis. Crucially, it is programmed to never provide actual diagnoses, treatments, or medical advice, instead focusing on identifying symptom patterns and flagging cases that require human follow-up. It operates under strict rules to ensure safety and prevent misuse, always responding in a structured JSON format.
Medication Agent
The Medication Agent offers non-prescription medication suggestions and lifestyle recommendations for mild conditions. Unlike many existing systems that provide opaque outputs, this agent emphasizes explainability by detailing the reasoning behind its recommendations. It strictly avoids recommending prescription drugs or controlled substances and is designed to escalate any serious or ambiguous symptoms for clinician evaluation, prioritizing patient safety.
Also Read:
- CLARITY: An AI Platform for Streamlined Patient Routing and Clinical Consultations
- AI Agents Accelerate New Alloy Development for 3D Printing
Appointment Agent
This agent handles patient scheduling, updates, cancellations, and medical history management. It interprets requests related to appointments and structures them for administrative purposes. Role-based policies strictly govern access to these functions, ensuring that only authorized individuals (patients, providers, staff) can initiate or modify appointments, aligning with best practices in health informatics.
A standout feature of the Agentic-AI Healthcare platform is its dedicated Privacy & Compliance Layer. This layer is meticulously designed to align with major healthcare data protection standards, including HIPAA (US), PIPEDA (Canada), and PHIPA (Ontario). Key components of this layer include:
- AES-GCM Field-Level Encryption: Patient health information is encrypted at a granular level, significantly reducing re-identification risks.
- Role-Based Access Control (RBAC): Permissions are carefully allocated based on user roles (e.g., patient, doctor, auditor) to ensure that individuals only access the information necessary for their tasks, adhering to the principle of least privilege.
- Tamper-Evident Audit Logging: All access and modifications to encrypted records are meticulously logged in tamper-evident hash chains, providing a trustworthy record for accountability and compliance monitoring.
- Consent and Transparency: Access to personal health information is mediated by explicit patient consent, reinforcing ethical data handling practices.
The system also boasts multilingual accessibility, supporting interactions in English, French, and Arabic. This feature is particularly vital in diverse regions, ensuring that language barriers do not hinder access to digital health services. The input language is detected, and requests are routed to language-specific prompt templates, maintaining consistent output schemas across languages.
While presented as a research prototype and not a certified medical device, Agentic-AI Healthcare offers a compelling blueprint for the future of trustworthy healthcare AI. It demonstrates the feasibility of combining agentic orchestration, multilingual accessibility, and a compliance-aware architecture. Future work will focus on strengthening agent-level trust through authentication mechanisms, ensuring that only verified agents participate in workflows.
For a deeper dive into the technical details and methodology, you can read the full research paper here.
