Mira AI is an advanced healthcare provider assistant designed to empower medical teams in managing chronic conditions through a multi-agent AI architecture. By leveraging Large Language Models (LLMs) and clinical logic, Mira AI helps healthcare workers monitor patient health, manage treatment adherence, and coordinate interventions efficiently.
Mira uses a T-MAC (Team Multi-Agent Collaboration) architecture where specialized AI agents work together like a clinical team:
┌─────────────────────────────────────────────────────────────────────┐
│ MORNING ROUNDS WORKFLOW │
├─────────────────────────────────────────────────────────────────────┤
│ │
│ 1️⃣ LOAD DATA │
│ └── Fetch all active patients for the organization │
│ │
│ 2️⃣ SPECIALIST ANALYSIS (Parallel) │
│ ├── 🔬 HIV Specialist → Checks VL, adherence, regimen │
│ ├── 💊 Hypertension → Analyzes BP readings, medications │
│ ├── 📊 Engagement → Flags inactive patients │
│ └── 📅 Follow-Up → Schedules appointment reminders │
│ │
│ 3️⃣ SUPERVISOR SYNTHESIS │
│ └── Head Doctor merges all recommendations, resolves conflicts │
│ │
│ 4️⃣ AUTO-EXECUTION │
│ └── Low-risk actions (reminders, nudges) sent automatically │
│ │
│ 5️⃣ CRITIC REVIEW │
│ └── Quality Doctor validates high-risk clinical decisions │
│ │
│ 6️⃣ PERSIST & COMPLETE │
│ └── All actions saved with full decision trace for audit │
│ │
└─────────────────────────────────────────────────────────────────────┘
| Agent | Role | Key Capabilities |
|---|---|---|
| Head Doctor (Supervisor) | Orchestrates morning rounds | Coordinates specialists, synthesizes decisions, handles conflicts |
| HIV Specialist | ART management expert | VL monitoring, defaulter tracing, regimen optimization |
| Engagement Specialist | Patient activity monitor | Flags disengaged patients, sends onboarding reminders |
| Follow-Up Specialist | Appointment coordinator | Smart reminder scheduling based on no-show history |
| Clinic Critic | Quality control | Reviews AI decisions before execution |
Watch the AI "think" in real-time! Mira streams each agent's reasoning process via Server-Sent Events (SSE):
# 1. Trigger analysis
curl -X POST "http://localhost:8000/api/v1/agents/trigger-rounds" \
-H "Authorization: Bearer TOKEN"
# Response includes cycle_id
# {"data": {"cycle_id": "abc-123..."}}
# 2. Subscribe to thought stream
curl -N "http://localhost:8000/api/v1/agents/stream/abc-123" \
-H "Authorization: Bearer TOKEN"Example Stream Output:
🏥 Beginning morning rounds for organization...
📋 Loaded 15 active patients for analysis
🔬 Starting HIV review for 8 HIV patients...
⚠️ John Doe: VL=1500 copies/ml (UNSUPPRESSED). Requires urgent counseling.
✅ Jane Smith: VL<50 (UNDETECTABLE). Excellent response!
📊 HIV Review complete: 8 patients analyzed, 3 interventions proposed.
📝 Synthesis complete. 17 final actions decided.
⚡ Executing automated low-risk actions...
🔍 Critic reviewing all proposed actions for safety...
✅ Morning rounds complete. 17 actions ready.
- Organization Management: Built for hospitals, clinics, NGOs as distinct tenants
- Role-Based Access Control (RBAC): Secure access for Doctors, Nurses, and Coordinators
- Audit Logging: Comprehensive tracking of all authentication and clinical events
- Modular Condition Profiles: Extensible architecture for disease-specific data
- Automated Defaulter Tracing: Flags patients who miss refills or appointments
- Action Idempotency: 48-hour cooldown prevents duplicate notifications
- Decision Trace: Full audit trail of AI reasoning for compliance
- Celery Background Workers: Handles email delivery and periodic analysis
- Email/SMS Notifications: Automated patient reminders and nudges
| Category | Technology |
|---|---|
| Framework | FastAPI (Python 3.12+) |
| Database | PostgreSQL + SQLAlchemy |
| AI/LLM | Google Gemini |
| Streaming | Redis Pub/Sub + SSE |
| Tasks | Celery |
| Infrastructure | Docker |
- Docker & Docker Compose
- Google Gemini API Key
# 1. Clone the repository
git clone https://github.com/gemini-hack/backend.git
cd backend
# 2. Configure environment
cp .env.sample .env
# Edit .env with your GOOGLE_API_KEY, DATABASE_URL, REDIS_URL
# 3. Start the stack
docker compose up --build
# 4. Access the API
# API: http://localhost:8000/api/v1
# Docs: http://localhost:8000/docs├── alembic/ # Database migrations
├── app/
│ ├── agents/ # Multi-agent AI system
│ │ ├── supervisor.py # Head Doctor orchestration
│ │ ├── workers/ # Specialist agents (HIV, Engagement, etc.)
│ │ ├── context.py # Shared agent state
│ │ └── thought_emitter.py # Real-time streaming
│ ├── api/ # FastAPI routes
│ ├── models/ # SQLAlchemy models
│ ├── schemas/ # Pydantic validation
│ ├── services/ # Business logic
│ │ └── action_idempotency.py # Duplicate prevention
│ └── workflows/ # Background workflows
├── docs/
│ └── images/ # Architecture diagrams
├── Dockerfile
└── compose.yml
To add a new disease specialization:
- Create a profile model in
app/models/conditions.py - Add a specialist agent in
app/agents/workers/ - Register the worker in
app/agents/factory.py - Add thought emissions for real-time streaming
Distributed under the MIT License. See LICENSE for more information.