Step 1: Create and Customise Personas
Step 2: Set Up a Focus Group
Step 3: Watch AI Agents Interact and Converse
Step 4: Review Focus Group Outcome and Analytics

MeshAI - AI-Powered Focus Group Platform

An innovative platform that transforms traditional market research. Demo use case focuses on AI-powered focus groups, enabling businesses to gather authentic insights from AI agents with customized personas that simulate real customer behaviors and perspectives.

Inspiration

The inspiration for MeshAI came from wanting to simulate the human community using AI agents, or building a community of AI agents, which we believed had such a huge scope of potential application. We decided to zoom in to a very specific use case of focus groups due to the challenges businesses face in conducting traditional focus groups - high costs, scheduling difficulties, limited participant diversity, and geographic constraints. We envisioned a world where companies could instantly access diverse perspectives from AI personas that authentically represent different demographics, psychographics, and behavioral patterns.

Our vision was to democratize market research by making high-quality consumer insights accessible to businesses of all sizes, from startups to enterprises, while maintaining the authenticity and depth of human feedback through advanced AI simulation.

What it does

MeshAI allows anyone to simulate of AI communities. In the prototype use case, we use it to power focus groups that enables businesses to conduct the following:

🎭 AI Persona Management

Create and customize AI personas with detailed backgrounds, demographics, and behavioral traits
Choose from pre-built personas including industry experts, consumer archetypes, and notable figures
Each persona maintains consistent personality, opinions, and decision-making patterns across sessions

🗣️ Interactive Focus Groups

Conduct real-time focus group sessions with multiple AI personas
Simulate authentic group dynamics with natural conversation flow
Generate streaming responses that mimic real human interaction patterns

📊 Advanced analytics

Real-time sentiment analysis of participant responses
NPS (Net Promoter Score) and CSAT (Customer Satisfaction) metrics
Comprehensive reporting with insights extraction and trend analysis

🎯 Multi-Modal Research

Simple Q&A interactions for quick feedback
Group discussions with dynamic persona interactions
Structured focus group sessions with defined goals and outcomes

How we built it

Architecture & Tech Stack

Frontend (React + TypeScript)

Framework: React 18 with TypeScript for type safety
Build Tool: Vite for fast development and optimized builds
UI Components: Radix UI primitives with shadcn/ui design system
Styling: Tailwind CSS for responsive, modern design
State Management: React hooks with custom API client
Routing: React Router DOM for navigation

Backend (Python + Flask)

Framework: Flask with CORS support for API endpoints
AI Orchestration: CrewAI for managing multiple AI agents
LLM Integration: Google Gemini 2.5 Flash via LangChain
Configuration: YAML-based agent and task management
Data Storage: JSON-based persona storage with file system management

AI & ML Components

CrewAI: Multi-agent AI framework for persona simulation
Google Gemini: Advanced language model for natural conversations
LangChain: LLM integration and prompt management
Custom Sentiment Analysis: Real-time emotion and satisfaction scoring

Key Implementation Details

Multi-Agent AI System: Each persona is implemented as a separate CrewAI agent with unique:
- Role definitions and goals
- Background stories and motivations
- Consistent personality traits
- Memory across conversations
Real-time Communication:
- RESTful API endpoints for seamless frontend-backend communication
- Streaming responses for natural conversation flow
- Error handling and fallback mechanisms
Persona Engine:
- Dynamic persona loading from JSON configurations
- Customizable persona creation with avatar selection
- Persistent persona storage and retrieval
Analytics Pipeline:
- Real-time sentiment analysis using AI-powered text analysis
- Automatic NPS and CSAT score generation
- Comprehensive session analytics and reporting

Challenges we ran into

1. Multi-Agent Coordination

Managing multiple AI personas in a single conversation while maintaining individual personality consistency was complex. We solved this by:

Implementing careful prompt engineering for each persona
Using CrewAI's task orchestration to manage conversation flow
Creating memory systems to maintain context across interactions

2. Real-time AI Response Generation

Generating authentic, real-time responses from multiple AI agents simultaneously presented performance challenges:

Optimized LLM API calls with proper rate limiting
Implemented response caching and streaming
Built fallback mechanisms for API failures

3. Sentiment Analysis Accuracy

Creating accurate sentiment analysis that captures nuanced emotional responses:

Developed custom sentiment scoring algorithms
Integrated multiple AI models for cross-validation
Fine-tuned analysis parameters for marketing context

4. Dependency Management

Managing complex Python dependencies with LangChain, CrewAI, and Google AI:

Resolved version conflicts between packages
Created isolated virtual environments
Implemented comprehensive error handling

5. User Experience Design

Designing an interface that makes AI-powered focus groups feel natural and intuitive:

Iterative UI/UX testing and refinement
Real-time visual feedback for AI responses
Responsive design for various screen sizes

Accomplishments that we're proud of

🚀 Technical Achievements

Seamless Multi-Agent AI Integration: Successfully orchestrated 20+ AI personas with distinct personalities
Real-time Conversation Engine: Built streaming AI responses that feel natural and engaging
Advanced Analytics Dashboard: Created comprehensive insights extraction from AI conversations
Scalable Architecture: Designed a system that can handle multiple concurrent focus group sessions

💡 Innovation

AI Persona Authenticity: Achieved remarkable consistency in persona behavior across different scenarios
Dynamic Group Interactions: Enabled AI personas to interact with each other, not just respond to prompts
Sentiment Intelligence: Developed sophisticated emotion and satisfaction analysis for market research

🎨 User Experience

Intuitive Interface: Created a user-friendly platform that requires no AI expertise
Professional Analytics: Built enterprise-grade reporting and insights visualization
Customization Freedom: Enabled users to create and modify personas for specific research needs

⚡ Performance

Fast Response Times: Optimized AI response generation for real-time conversations
Reliable System: Implemented robust error handling and fallback mechanisms
Scalable Infrastructure: Built a foundation that can grow with user demand

What we learned

AI Development Insights

Prompt Engineering is Critical: The quality of AI persona responses heavily depends on carefully crafted prompts and context
Multi-Agent Complexity: Managing multiple AI agents requires sophisticated orchestration and memory management
LLM Integration Challenges: Working with different AI models requires understanding their unique capabilities and limitations

User Experience Design

AI Interface Design: Learned how to make AI interactions feel natural and trustworthy
Data Visualization: Discovered effective ways to present complex analytics in digestible formats
Responsive Design: Ensured the platform works seamlessly across devices

What's next for Mesh AI

🌟 Long-term Vision

Global Platform

International Expansion: Support for global markets with localized personas
Compliance Framework: GDPR, CCPA, and other privacy regulation compliance
Mobile Applications: Native iOS and Android apps for on-the-go research

AI Innovation

Hybrid Human AI Groups: Combine AI personas with real human participants
Virtual Reality Integration: Immersive focus group experiences
Emotional AI: Advanced emotion recognition and response generation

Market Expansion

Academic Research: Tools for universities and research institutions
Government Applications: Public policy research and citizen engagement
Non-profit Sector: Accessible research tools for social impact organizations