Nexus | Devpost

Inspiration

Imagine, at the beginning of this weekend's event, we put all 1,800 hackers in one room and asked everyone to make a new friend.

What are the odds that you meet someone incredible? Not too bad.

But to discover and cultivate a genuine connection, it takes more than just luck. And why leave it to chance?

Introducing Nexus, graph visualization software that helps you discover meaningful relationships.

What it does

One of the most hectic parts of a hackathon is always team formation. TreeHacks 10 saw a record number of hackers, and with it, came a #team-searching Slack channel like never before. Messages upon messages of overachievers and lost souls, total beginners mixed with those who can code with eyes closed. Within all this mess, how could you possibly find a team?

Nexus takes the data from these messages and creates a dynamic, interactive 3D knowledge tree of this year's incredible cohort of students, showcasing their interests, skills, and backgrounds. It uses semantic abstraction to let you search and explore this network to find and start conversations with those who have similar ideas.

How we built it

Parsing messages from TreeHacks' Slack channels, we scraped data and used Together AI to generate embeddings and perform NER (Named Entity Recognition), identifying the interests and background of each hacker.

For vector search, we tried both Convex and Chroma DB. This allowed us to embed and quantifiably compare each individual using cosine similarity and consider a link between two people only if the value is above a given threshold. The 3D graphs were rendered with a framework built on Three.js, creating a spatial and interactive way of exploring connections.

Challenges we ran into

One of the most significant challenges was ensuring the accuracy and relevancy of data extracted from Slack. The diversity of conversation formats and the informal language used required fine-tuning our models for better NER and embedding generation.

Another challenge was optimizing the 3D visualizations for performance without compromising on detail. We built on top of Three.js, which made visualizing nodes and edges extremely quick.

Accomplishments that we're proud of

Committing to a project after 17 hours of brainstorming and making prototypes for ideas that did not end up coming to life

What we learned

How to take abstract and noisy data and produce high-quality comparisons between individual interests

What's next for Nexus

Creating more complex visual representations of graph-based entities for commercial use (e.g., mapping LinkedIn networks for founders/investors, generating supply chain graphs for large-scale manufacturing)