Abstract

Mental health challenges have become a global concern, with millions experiencing stress, anxiety, and depression. However, immediate support and intervention are often inaccessible. This project presents Mental Health Companion Agent (MHA) — an AI-powered autonomous agent system designed to detect emotional distress from natural language conversations and provide personalized, empathetic support along with resource recommendations.

The MHA system leverages multi-agent collaboration (built using CrewAI) where agents specialize in sentiment analysis, contextual conversation generation, and resource curation. Using RoBERTa for emotion detection, Mistral 7B for dynamic response generation, and an easy-to-use Gradio interface, the system creates a seamless, interactive mental health companion.

This project demonstrates how Agentic AI can empower mental health support systems, aligning directly with the challenge's theme — showcasing practical, real-world impact through autonomous AI workflows.

Introduction

The global mental health crisis is one of the most pressing societal challenges today. According to WHO, over 280 million people globally suffer from depression, and access to professional help is limited by availability, stigma, and costs.

Technology can bridge this gap. Conversational AI combined with emotional intelligence detection has the potential to provide first-line support, guiding individuals towards self-help resources or professional care if needed.

This project aims to build an AI companion capable of detecting distress and offering personalized support, powered by Agentic AI workflows that bring together specialized agents for emotion analysis, response generation, and recommendation delivery.

Methodology

System Overview
The project follows a multi-agent architecture powered by CrewAI, where each agent specializes in a core task:

System Overview
Agent Name Role
🧠 Sentiment Analysis Agent Analyzes the
incoming message using RoBERTa to detect emotional tone (e.g., anxiety, sadness, stress).
💬 Response Generation Agent Generates an empathetic response using Mistral 7B, aligning with detected emotions.
📚 Resource Recommendation Agent Curates and suggests mental health resources (self-help articles, breathing exercises, helplines, etc.) relevant to the detected emotional state.

Technical Components
Component Technology Used
LLM Mistral-7B
Emotion Analysis RoBERTa (Fine-tuned for sentiment/emotion detection)
Agent Framework CrewAI
UI Gradio
Orchestration Python-based CrewAI scripts

Workflow Diagram

Experiments

Sample Test Cases
User Input Detected Emotion Generated Response Recommended Resource
I feel really anxious about my exams. Anxiety I hear you, exams can be overwhelming. Take a deep breath — here are a few stress-relief techniques that might help. Breathing exercises, exam stress tips
I feel alone and helpless lately. Sadness I'm really sorry you're feeling this way. It might help to talk to someone you trust — here are a few helpline numbers and self-help guides. Helpline links, coping strategies

Results

System Performance
Metric Result
Sentiment Detection Accuracy (RoBERTa) ~89%
Response Coherence (Mistral-7B) High (based on manual evaluation)
Resource Relevance Curated list based on expert-reviewed resources
Key Observations
Multi-agent orchestration works smoothly, with clear task division ensuring modular and scalable design.
Sentiment analysis is highly accurate, especially for distress signals.
Mistral-7B generates responses that feel natural and empathetic.
Combining conversation + resources creates a holistic support experience.

Conclusion

The Mental Health Companion Agent demonstrates how Agentic AI can contribute meaningfully to addressing real-world challenges, specifically in mental health support.

This project serves as an autonomous first-line companion that can provide timely, empathetic, and personalized support to individuals in distress — a capability highly relevant in today’s mental health crisis landscape.

By combining NLP advancements, autonomous agent frameworks, and user-friendly interfaces, this project not only showcases technical prowess but also highlights the potential of AI for social good, making it a strong contender in the Agentic AI Innovation Challenge 2025.