ConnectMax

Professional networking today is fundamentally broken. Platforms like LinkedIn connect people, but they don't optimize for actual career growth or provide strategic pathways for professional development. We wanted to create something that doesn't just connect people, but actually computes the optimal path for each person's career growth using advanced AI. The inspiration came from asking: "What if we could treat career development like a mathematical optimization problem, where each connection and interaction is strategically chosen to maximize someone's chances of success?"

ConnectMax is an AI-powered professional networking platform that uses active inference (the computational framework that models how the human brain makes decisions) to create personalized networking strategies. The system works in two key phases:

1. Profile Building Phase:

   • An AI agent engages users in adaptive conversations
   • Uses active inference to dynamically select questions that will maximize information gain
   • Adjusts its questioning strategy based on user responses, resistance to topics, and engagement levels
   • Autonomously decides when it has gathered enough information to build a comprehensive profile

2. Strategy Generation Phase:

   • Generates detailed profiles of 5 specific people the user should connect with
   • Creates a strategic progression path, detailing when and how to approach each connection
   • Provides specific objectives and readiness indicators for moving between connections
   • Explains why each connection is valuable for the user's goals

How we built it The system is built on three main technical components:

1. Active Inference Engine:

   • Implemented using the PyMDP framework
   • Maintains belief states about user traits, goals, and interests
   • Computes Expected Free Energy for potential questions
   • Uses Dirichlet learning to update its observation models

2. Multi-Agent System:

   • Profile Builder Agent: Handles the questioning process
   • Concept Map Manager: Analyzes conversation history to build comprehensive profiles
   • Community Matcher: Generates strategic networking recommendations

3. Frontend:

   • Built with Next.js and Tailwind CSS
   • Real-time chat interface with adaptive response handling
   • Social media-style display of recommended connections
   • Strategic pathway visualization

Challenges we ran into

1. Active Inference Implementation:

   • Balancing exploration vs. exploitation in question selection
   • Tuning the social cost parameters for different question types
   • Handling the computational complexity of Expected Free Energy calculations

2. User Experience:

   • Making the AI's questioning process feel natural and conversational
   • Designing an interface that makes complex recommendations easily digestible
   • Balancing detail vs. clarity in the strategic progression path

3. Technical Integration:

   • Coordinating multiple AI agents while maintaining conversation coherence
   • Managing state transitions between profile building and recommendation phases
   • Optimizing response times for real-time interactions

Accomplishments that we're proud of

1. Successfully implemented a brain-inspired AI system that adapts its questioning strategy based on user responses
2. Created a unique approach to professional networking that gets exponentially more valuable as the user base grows
3. Developed a sophisticated multi-agent system that generates actionable, personalized networking strategies
4. Built a privacy-conscious architecture that protects sensitive user information
5. Created an intuitive interface that makes complex AI recommendations accessible and actionable

What we learned

1. The power of active inference in creating truly adaptive AI systems
2. The importance of balancing technical sophistication with user experience
3. How to handle sensitive user data responsibly while maintaining system functionality
4. The complexities of building multi-agent systems that work coherently together
5. The challenges of implementing brain-inspired computing in real-world applications

What's next for ConnectMax

1. Technical Enhancements:

   • Extend the active inference model to plan multiple questions ahead
   • Implement speech analysis to detect subtleties like sarcasm and uncertainty
   • Add multi-modal input processing (voice, text, etc.)

2. Feature Expansion:

   • Develop a group recommendation system for team building
   • Create an API for integration with other professional development tools
   • Implement real-time connection availability tracking

3. Privacy and Security:

   • Plan Implement full end-to-end encryption for user profiles
   • Plan to develop advanced anonymization techniques for network-level recommendations
   • Plan to create user-controlled data sharing preferences

4. Platform Growth:

   • Build partnerships with professional organizations
   • Develop industry-specific recommendation models
   • Create an enterprise version for corporate talent development

Our ultimate goal is to make ConnectMax the standard platform for strategic professional development, where career growth and community building is treated as a mathematical optimization problem that gets better at finding solutions with each new user interaction.

Built With

• langgraph
• nextjs
• pymdp
• python