mindSync

Inspiration

MindSync was inspired by the belief that genuine conversations foster deep insights and meaningful, lasting relationships. Traditional social networks often prioritize superficial interactions, which motivated us to build a platform that facilitates meaningful engagements. Research, such as "Friendships in Old Age: Daily Encounters and Emotional Well-Being" (Ng et al., 2020), emphasizes that even brief, positive encounters significantly enhance emotional well-being. MindSync leverages insights from personality psychology and social networking research to create authentic connections by grouping individuals with shared interests and compatible personality traits.

What it Does

MindSync connects like-minded individuals through thoughtful, weekly-themed questions designed to reveal their personalities, passions, and viewpoints. Responses are analyzed by our advanced algorithms, grouping users into chat rooms where genuine and meaningful conversations naturally unfold.

How We Built It

MindSync integrates several cutting-edge technologies:

Frontend: Built with Next.js, ensuring a dynamic, responsive, and intuitive user experience for both desktop and mobile users.
Backend: Utilized Supabase for secure and efficient data management and authentication.
Personality Matching: Implemented personality matching using Personality_LM, inspired by research such as "Continuous Output Personality Detection Models via Mixed Strategy Training" (Wang & Sun, 2024). This model categorizes users based on psychological theories, enabling deeper, personality-based connections.
Grouping Algorithm: Developed a Genetic Algorithm inspired by studies on evolutionary algorithms for clustering (Goldberg, 1989; Zhao et al., 2019). This algorithm maximizes intra-group cohesion by aligning users with compatible interests and personalities.

Tech Stack

Frontend: Next.js
Backend: Supabase
Algorithm API: FastAPI
Machine Learning: PyTorch and GATv2Conv, informed by research on "Graph Attention Networks" (Velickovic et al., 2018), refining user embeddings to accurately reflect group cohesion.
Real-Time Chat: Implemented seamless real-time communication using WebSockets and FastAPI for smooth, low-latency user interactions.

Challenges We Ran Into

Grouping Algorithm Tuning: Balancing personality compatibility and optimal group sizes proved challenging, requiring careful tuning of our genetic algorithm.
Real-Time Chat: Ensuring reliable, low-latency real-time chat functionality demanded consistent optimization.
Performance and Scalability: Designing a robust backend capable of scaling efficiently was critical, necessitating thorough planning and performance testing.
Consistent Styling Configurations: Properly integrating intuitive, accessibility-friendly styling was a major challenge which led to us adopting a simple purple theme.
Huggingface Model Deployment: Navigating dependency conflicts, Heroku's resource limits, and the large size of our custom LLM presented significant deployment obstacles.

Accomplishments We’re Proud Of

Successfully refining user embeddings using Graph Neural Networks (GNNs).
Developing and optimizing a sophisticated Genetic Algorithm to maximize group cohesion and meaningful connections.
Seamlessly integrating robust real-time chat features.
Deploying Personality_LM effectively for accurate personality classification, significantly enhancing the depth of connections on MindSync.

What We Learned

This project underscored the importance of user-centered design, iterative testing, and continuous optimization of real-time systems.

Our exploration of personality psychology, especially the Big Five Personality Traits theory, informed critical design and functionality decisions. Additionally, understanding the significant positive impact of meaningful daily social interactions, as highlighted in research by Ng et al. (2020), shaped our approach to fostering genuine connections.

What's Next for MindSync

Future enhancements include:

Refining the Matching Algorithm: Incorporating advanced psychological insights to enhance user compatibility further.
Adding Gamification: Introducing features like badges, leaderboards, and interactive challenges to boost user engagement.
Expanding Communication Features: Including video chat capabilities and personalized topic channels for deeper interaction.
Analytics Dashboard: Developing tools to provide insights into community dynamics, further tailoring user experiences.

Our goal is to evolve MindSync into a more interactive platform that continues fostering meaningful, enriching connections.

Research Papers and Resources

"Continuous Output Personality Detection Models via Mixed Strategy Training" (Wang & Sun, 2024)
"Graph Attention Networks" (Velickovic et al., 2018)
"Genetic Algorithms in Search, Optimization, and Machine Learning" (Goldberg, 1989)
"A Genetic Algorithm for Dynamic Clustering" (Zhao et al., 2019)
"Real-Time Messaging: A Survey of the Architecture and Challenges" (Lee et al., 2020)
"Friendships in Old Age: Daily Encounters and Emotional Well-Being" (Ng et al., 2020)