Polymaths - Intelligent AI Experts, On Demand

Inspiration

As professionals, we struggle to access specialized expertise when they need it most — whether it's a startup founder needing tech strategy advice at 2 AM or an HR manager dealing with a mental health crisis. Generic AI tools lack the depth and personalization required for complex professional challenges.

What it does

Polymaths AI creates personalized AI experts in any domain with realistic video avatars. Users complete a questionnaire about their needs, and our system generates comprehensive knowledge bases using advanced AI research, then crafts custom AI personas that match their communication style and expertise requirements for natural video conversations.

How we built it

• Frontend: React/TypeScript with Tailwind CSS for responsive UI
  
• Backend: Supabase with PostgreSQL, Row Level Security, and serverless edge functions
  
• AI Research: Anthropic Claude 4 Sonnet for comprehensive knowledge generation
  
• Video Avatars: Tavus AI API integration for realistic AI persona conversations
  
• Architecture: Real-time database with secure authentication and scalable edge computing
  

Challenges we ran into

• Timeout Issues: Initial OpenAI Deep Research API calls exceeded edge function limits, requiring migration to faster Claude API . Also it costed a lot without any added benefits so had to switch.
• Knowledge Quality: Ensuring AI-generated expertise met professional standards across diverse domains
  
• Video Integration: Seamlessly connecting custom AI personas with Tavus video avatars while maintaining conversation context

Limited use of external tools

To fix certain issues, I was facing some issues in bolt and it was going into error loops so I had to use Claude Code to fix those. These are the issues fixed with external help:

• The edge function timeout issue when creating knowledge bases with deep research
• Logout functionality where users stayed on dashboard instead of being redirected to the landing page after signing out.
• Adding avatar images and landing page video to public folder, done manually in cursor and pushed to git. Was simple that way.

Accomplishments that we're proud of

• Successfully generated comprehensive, professional-grade knowledge bases as per Tavus AI avatar requirements for any domains
  
• Created a seamless knowledge generation workflow
• Ability to link any knowledge base to any personal / polymath
• Created seamless video conversation experiences with personalized AI experts
  
• Built a scalable architecture that can rapidly deploy new AI experts
  
• Developed pre-defined diverse polymaths covering mental health, tech innovation, and sustainability
  

What we learned

• Quality over Speed: Fast AI responses matter less than expertise depth and personalization
  
• User-Centric Design: The questionnaire-driven approach creates more relevant AI experts than file uploads
  
• Integration Complexity: Combining AI research, persona creation, and video avatars requires careful orchestration
  
• Domain Specificity: Users prefer specialized experts over generalist AI assistants
  

What's next for Polymaths - Intelligent AI Experts, On Demand

• Enterprise Integration: Direct integration with Slack, Teams, and corporate learning platforms
  
• Expert Marketplace: Community-contributed AI experts across niche professional domains
  
• Advanced Personalization: Learning from user interactions to continuously improve expert responses
• Multi-modal Expertise: Detailed knowledge generation pipeline, adding document analysis, presentation creation, and collaborative problem-solving capabilities
  
• Real experts Integration: Working with professional associations to create certified AI experts in regulated fields to provide a library of pre-defined real experts.

Built With

• anthropic
• bolt.new
• claude
• framer-motion
• radixui
• react
• supabase
• tailwindcss
• tavusai
• vite