friended. | Devpost

Inspiration

We were inspired to make this project after we went through the process of finding a hackathon team ourselves. Though there exists a Discord channel and Devpost's features to filter through participants, we felt that they were not enough to help people reliably create teams. We felt that there was a method with greater clarity that could be established to help team members come together all while keeping compatibility in mind. That's where the idea of friended. was born.

What it does

There is plenty of data about what makes you, you 😊. It’s just siloed in a bunch of apps 😔. That data is useful because it can help you find you’re next hackathon team, co-founder, friend, partner… friended. brings your digital data into one place and finds people like you. It takes your interests, skills, and experiences from Devpost and LinkedIn and leverages OpenAI embeddings to allow for Supabase vector similarity search. This allows for use cases like helping hackathon organizers set up teams. Organizers can set up events for participants to meet and discover people participating in the hackathon with similar interests, skills, and work experiences. Ultimately, friended. helps you build meaningful connections through your pre-existing, public data.

How we built it

The front end was designed using Next, TypeScript and Tailwind CSS. Moreover, the website was first designed in Figma before implementation.

The backend was designed using Supabase. We chose Supabase for its low-latency and its support for embeddings. We used web scrapers to scrape data from LinkedIn and Devpost. This data was then embedded using OpenAI's embedding tool. Then, PostgreSQL was used to apply cosine similarity to determine how closely related two people were through the use of pgvector.

Credit for the scrapers:

Devpost Scraper

LinkedIn Scraper

Challenges we ran into

Our primary challenge was determining what sites to scrape public user data from to use in cosine similarity. Not much data is made publicly accessible, so we had to find sites which we could actually obtain data from. After a long period of testing and experimenting, we were able to arrive at LinkedIn and Devpost. As a result, we made the application more computer science-oriented and geared it towards hackers and hackathons. However, its other use-cases are still possible. This is an area for future development.

Accomplishments that we're proud of

We are quite proud of the fact that we were able to learn and implement Supabase with minimal prior knowledge. It truly helped to organize the backend, making development a lot easier. We are proud that we were able to obtain another technology in our toolkits.

What we learned

Aside from learning new technologies such as Supabase and pgvector, we also learned the importance of contingency planning. Due to time constraints and intense setting, many things did not go as we originally planned. It was important for us to able to quickly shift gears and adapt to the situation.

What's next for friended.

Friended will not end with this hackathon. We plan to add more sites other than Linkedin and Devpost to our similarity program. Moreover, we want to reach out to hackathon organizers to possibly use our site to help hackers make teams.