The usual keyword-based search is very limiting.
We leveraged OpenAI's vector embeddings (Ada v2 model) along with Weaviate, a vector database, to provide a semantic search. We further factored in engagement metrics and time parameters to deliver relevant results.
Jam surfaces the most trending news on Farcaster by analyzing shared links using an engagement-oriented scoring mechanism that also takes into account their recency. It uses link metadata and a custom data-cleaning process to deliver clear, informative news content.
The Twitcaster feature uses AI-powered semantic search and Twitter APIs. This helps solve the cold start problem for new Farcaster users by suggesting relevant users and content based on their 50 recent tweets.
To show relevant content to the users, we analyze the most recent posts and categorize them into the top 20 topics by using openAI APIs and Davinci-003 Model.
As the first adopter of Farcaster Hubs, we worked very closely with the Farcaster protocol team and provided them with early feedback. By integrating hubs, we were able to create a smooth user experience.
We also made use of long casts to surpass the hub's character limits and switched to direct data sync from hubs, adeptly handling design constraints.
Using Terraform, Kubernetes, Docker, and AWS, our DevOps team facilitated quick product development and scalability. This additionally empowered the developers to create preview environments that enhanced collaboration and innovation.
1st
Most widely used third-party app on Farcaster
“Just discovered the search on Jam with filter by time. Way easier to find new stuff on jam with this. Great feature.”
-Fran (Farcaster user)90%
People who try Jam continue to use it at very high rates resulting into 90% wau/mau