Collide Content Recommendation Engine

What we made

Our prompt was to create a content ranking engine for Collide, a community-driven knowledge sharing platform for the energy industry. We were given raw posts, comments, fake user profiles, and we built a website with a recommendation engine for users and posts. It uses semantic searching with a vector database to find similarities among users and their content of interest.

How we did it

Using these technologies:

Frontend: React, SCSS
Backend: Flask, OpenAI
Database: Qdrant (for vector search)

We came up with an algorithm that collects all data related to each user. Their posts, likes, dislikes, and comments are all vectorized, weighted, and then semantic searched against all posts.

Biggest challenges

Having only one team member come up with an entire UI for such a complex project in so little time.
Designing an algorithm that's fast enough and accurate enough to be a proper recommendation engine.
Learning an entirely new database system that acts completely differently from traditional relational databases.

What we're proud of

We came up with an entire algorithm without any prior experience in designing recommendation engines.
One team member managed to create the UI and frontend for our project.
We split the work between backend and frontend effectively.

What's next

Same as the Hackathon, collecting as many unbiased ideas as possible to expand on creativity and come up with great additions for Collide is a great way to move forward.