Cabal Sorel

Cabal Sorel is a simple Chrome extension aiming to make web3 social media sites better by offering smarter suggestions. We've noticed an issue with web3 socials - it's pretty tough to jump from one platform to another, and the lack of shared info means good recommendation systems are missing. Our goal is to help web2 socials shake hands with web3. With our extension, you can keep track of what you do across many different platforms and get suggestions based on that. We have a smart suggestion system that learns from your interactions to make this happen.

Tell us about how you built this project; the nitty-gritty details. What technologies did you use? How are they pieced together? If you used any partner technologies, how did it benefit your project? Did you do anything particularly hacky that's notable and worth mentioning?

This project was built using a chrome extension as the primary user interface. This choice was made because it will allow Cabal Sorel to integrate with multiple sites with ease. The primary technology used for the chrome extension were the chrome extension manifest, html, javascript, css, index database, and service-workers. This combination of technologies allows for a secure way of processing the onboarding and makes the data management simple.

We use NextId to centralize all platform connections in one unified location. During the initial setup, we prompt users to link their accounts to NextId via the Proof Service API. This step creates a NextId account that becomes the nexus for all their social media profiles. Leveraging the capabilities of the Relational Service API, we can then effortlessly fetch data from all connected social platforms. This setup not only simplifies account management but also enables our extension to keep a comprehensive track of user interactions across the different platforms, fostering a more cohesive cross-platform social experience.

To construct the recommendation system, we leveraged the public BigQuery dataset provided by Lens to extract comprehensive data on likes, user profiles, and posts. We harnessed the power of Apache Spark to train an advanced ALS recommendation model capable of predicting user post preferences. Subsequently, we developed an API designed to seamlessly deliver a curated feed of post recommendations tailored to individual user preferences.

To maintain a decentralized record of web2 social interactions, we use Tableland as our database. This setup captures and preserves interactions, setting the stage for subsequent algorithm training. Additionally, Tableland empowers users by granting them ownership of their contributions towards refining our algorithms. This not only ensures data transparency but also cultivates a sense of user involvement in enhancing the recommendation engine over time.