NimaAI

Inspiration

Mental health support is often inaccessible and intimidating. NimaAI breaks down these barriers with 24/7 AI-powered support that orchestrates multiple specialized agents to provide comprehensive care—from crisis intervention to therapist matching to peer support.

What Was Learned

Multi-Agent Orchestration: Coordinating 6 specialized AI agents with smooth handoffs and state management
Conversational Design: Balancing empathy and intelligence using Gemini 2.5 Pro and Flash models
Voice Interface: Handling speech-to-text and text-to-speech with browser limitations and conversation flow
Privacy Architecture: Designing a 4-tier privacy system giving users full control
Cloud Deployment: Deploying complex AI systems on Google Cloud Run with auto-scaling

How It Was Built

Tech Stack:

FastAPI backend with 6 AI agents (Intake, Crisis, Resource, Habit, Support Group, Coordinator)
Google Gemini 2.5 (Pro for reasoning, Flash for speed)
Voice interface using WebKit Speech API
JSON database with 20 therapists and 11 support groups
Deployed on Google Cloud Run

Key Features:

Natural voice conversations with real-time speech recognition
Interactive booking modals for counselor selection, privacy, and scheduling
Smart matching for therapists across 9 specializations and peer support groups
Habit tracking with streaks and evidence-based recommendations
Real-time analytics dashboard

Challenges Overcome

Agent Loops: Implemented state flags to prevent agents from repeating questions
Modal Coordination: Synchronized backend state with frontend UI triggers
Speech Limits: Built chunking system for long text-to-speech responses
State Management: Created centralized AgentState for 6-agent coordination
CSS Conflicts: Resolved modal persistence issues with specificity overrides
Model Selection: Optimized speed versus intelligence across different agents

What Makes It Special

NimaAI is truly agentic, with each agent having autonomy and specialized knowledge, collaborating without rigid scripts. The voice interface and empathetic design prioritize accessibility, while the 4-tier privacy system ensures user control. It demonstrates how multi-agent AI can provide sophisticated, personalized mental health support accessible to anyone, anywhere, anytime.