Moha Health - Multi-Agent AI Health Assistant

Moha Health is a multi-agent AI healthcare assistant. Patients describe symptoms via voice or text, and AI agents triage cases, route them to specialists, and generate clinical reports in minutes.

Comment

Moha Health

My own health assistant. AI-powered intake that actually listens.

Built for Hack Canada 2026.

Inspiration

In Canada, healthcare often begins with a long wait.

Patients may spend hours waiting before anyone even asks detailed questions about their symptoms. The intake process — the first step in care — is often rushed, even though it plays a critical role in identifying urgent cases.

We built Moha Health to explore a simple idea:

What if the first step of healthcare was handled by an AI system that carefully listens, asks the right questions, and routes patients to the right care faster?

Instead of static forms or basic chatbots, Moha Health simulates a coordinated healthcare intake team made of specialized AI agents.

It doesn't replace doctors.
It simply helps patients reach the right care faster and more clearly.

What it does

Moha Health is a multi-agent AI healthcare assistant that simulates the first stage of a hospital visit.

Users describe symptoms using text or voice, and the system guides them through a structured intake process.

The system:

  1. Collects symptoms through a conversational AI intake nurse
  2. Routes cases to the appropriate specialist AI (dermatology, dental, cardiology)
  3. Asks focused follow-up questions
  4. Generates a clinical-style triage report

Optional inputs can enhance the assessment:

  • Symptom images
  • A short face video to estimate heart rate and respiration
  • A stored health profile for returning users

The result is a clear triage summary that can help identify urgency and guide the next step in care.

How it works

Moha Health uses a multi-agent AI architecture that mimics a hospital intake workflow.

The system pipeline can be summarized as:

$$ Patient \rightarrow Intake\ Nurse \rightarrow Router \rightarrow Specialist\ Agent \rightarrow Triage\ Engine \rightarrow Clinical\ Report $$

Each agent has a specific responsibility:

  • Intake nurse agent gathers symptoms and missing information
  • Router agent decides whether a specialist should be consulted
  • Specialist agents ask focused follow-up questions
  • Triage engine generates an urgency level and report

This modular design makes the system easier to extend with additional specialties in the future.

Key Features

Conversational Intake

An AI intake nurse collects symptom information such as location, severity, and duration.

Smart Routing

A routing agent determines whether a specialist consultation is needed.

Specialist Follow-up

Dermatology, dental, or cardiology agents ask targeted questions.

Triage + Clinical Report

The system generates a readable triage report and structured output.

Voice Interaction

Users can speak instead of typing using speech-to-text and text-to-speech.

Vitals from Video

A short face video can estimate heart rate and respiration.

Symptom Images

Users can attach a photo when describing visual symptoms.

Health Profile

Optional patient profiles allow returning users to store health history.

Tech Stack

Backend

  • FastAPI (Python 3.11)
  • Google Gemini via Backboard for LLM orchestration
  • Presage for vitals estimation from video
  • ElevenLabs for voice synthesis
  • Cloudinary for media uploads

Frontend

  • React
  • TypeScript
  • Vite
  • Tailwind
  • shadcn/ui

Deployment

  • Railway for backend deployment
  • Replit for frontend hosting
  • Tailscale for secure networking when needed

Challenges we ran into

Designing multi-agent coordination

The biggest challenge was building a system where multiple AI agents could collaborate while maintaining a coherent conversation with the user.

Careful prompt design and structured context passing were required to ensure that specialists received the right information.

Balancing AI reasoning with reliable triage

Healthcare applications require predictable outcomes.

To maintain reliability, Moha Health combines:

  • LLM reasoning
  • rule-based triage logic
  • structured report generation

This ensures results remain interpretable and consistent.

Integrating video vitals

Extracting heart rate and respiration from video required integrating external models and building a reliable media processing pipeline.

Handling uploads, processing, and responses smoothly during live interaction was a significant engineering challenge.

What we learned

Building Moha Health taught us several important lessons:

  • Multi-agent AI systems can mirror real-world workflows effectively
  • Healthcare AI must balance flexibility with safety and interpretability
  • Voice, vision, and language models together create more natural interactions
  • User experience is just as important as model capability

Most importantly, we learned that AI can assist healthcare workflows without trying to replace clinicians.

What's next for Moha Health

Future improvements could include:

  • more medical specialties
  • integration with clinical guidelines
  • additional physiological signals
  • multilingual support
  • integration with healthcare systems

The long-term goal is to create a digital front door to healthcare that helps patients reach the right care faster.

Sponsors & Thanks

This project was built during Hack Canada 2026, part of the Major League Hacking 2026 season.

Special thanks to:

Google Gemini
Tailscale
Stan
Cloudinary
Backboard
ElevenLabs
GitHub

Author

Nafisat Ibrahim

Website
LinkedIn
GitHub

Built With

Share this project:

Updates