The Problem
Mental health care has a supply-demand crisis. Over 970 million people live with a mental health disorder globally, yet the median treatment gap reaches 70% in low-income countries. In the US, the average wait time to see a psychiatrist is 25 days.
Our Solution
MeDo Mental Health Agent is a conversational AI companion built on three coordinated agents: an empathy layer, a therapeutic response engine, and a crisis detection module.
Core Features
Conversational Mental Health Assistant (POST /chat)
The empathy_agent analyzes messages for tone and distress signals. The therapeutic_response_agent selects the appropriate clinical technique:
- Grounding exercises for acute distress
- Psychoeducation for users seeking to understand symptoms
- Reflective listening for processing complex emotions
Mood Tracker (POST /mood-tracker)
The mood_insight_agent processes mood scores, emotion tags, sleep hours, and activity against evidence-based intervention logic.
Crisis Support (POST /crisis-support)
The crisis_assessment_agent scans messages for high-risk language. Based on severity:
- High: immediate crisis line routing
- Medium: warm handoff with counselor resource
- Low: supportive presence with continued conversation
Technical Stack
- Backend: FastAPI + Python asyncio
- Validation: Pydantic v2
- Concurrency: asyncio.gather() for parallel agent execution
- LLM Backend: Modular — OpenAI, Anthropic, or local models via Ollama
- Safe Messaging: Hardcoded guardrails independent of LLM output
Clinical Alignment
Grounded in CBT, DBT-informed grounding, Motivational Interviewing, and Exercise Psychiatry.
Privacy by Design
No persistent storage, ephemeral session IDs, no PII required, offline-capable with local LLM support.
Built With
- ai-agents
- asyncio
- crisis-detection
- fastapi
- mental-health
- natural-language-processing
- pydantic
- python
Log in or sign up for Devpost to join the conversation.