In this article we’ll highlight the Dash mini app and explain how it leverages ❜embed to deliver personalized content recommendations to each user.
Dash is a TikTok-like mini app that curates personalized short-form video content sourced directly from the Farcaster network. It was built and shipped by @benbassler.eth in May. Ben is also an engineer at weponder.io, a game of social intuition built on top of the Farcaster social graph, focusing on Dash on nights and weekends.
Dash’s feed personalization is ensured by AI models built and ran with the ❜embed platform. Each recommendation is based on a user’s social connections and interests mixed with discovery content. Let's now see how Ben has been using ❜embed to build this experience.
Don't forget to follow Dash’s development journey and regular updates on the /dash Farcaster channel.
How Ben Leveraged Embed’s Feed Builder
“I played around with the Embed feed builder for about 30-60 minutes, and I was ready to go ahead.”
— Ben Bassler
To deliver high-quality, personalized feeds efficiently, Ben chose the “For You” template available in the getembed.ai console and customized it to meet Dash’s specific needs. Here’s a breakdown of his configuration:
Detailed Feed Configuration
Feed Information
Feed Name: Dash For You Feed
API Endpoint: Currently using the For You endpoint (soon all endpoints will be unified into a single API).
1. Candidate Generation
Time period: Candidate generation window to a one-year period due to limited available video content.
Author IDs: Sources primarily include content from users the viewer follows and posts those users have interacted with.
Publication types: Explicitly filters for video content, while automatically removing spam and unsafe material.
2. Ranking
Scoring: Uses Embed’s balanced ranking algorithm, combining interest and social affinity scoring.
Time decay: Sets time decay to a low level, ensuring older yet engaging videos remain discoverable.
3. Visibility Filtering
Feed diversity: Applied diversity settings for enhanced feed variety:
Maximum posts per author: 2
Minimum distance between posts from the same author: 4
4. Feed Construction (Discovery Mix)
Promotions: Added a promotion block to integrate curated discovery content alongside personalized recommendations.
Dash Discovery: A distinct feed, Initially sourced discovery content from all users Ben followed, but is transitioning to a curated list of high-quality video creators to further enhance content relevance and reduce personal bias.
Percentage: Mixing the for you feed and discovery feed 50/50.
Each time a user loads their feed, Dash:
Calls the Embed API with the configured feed ID (additional parameters can be added as needed).
Fetches the personalized feed along with relevant metadata.
The feed is then ready to be served directly to users!
Features in the Pipeline for Dash and video feeds!
If you're building a video feed, we're cooking a few things to enhance video feeds that all builders will be able to leverage soon:
Advanced content analysis: Introducing automated video labeling for finer content filtering.
User interaction insights: Deepening the understanding of user engagement to further refine personalization.
How to get started:
-> No Code: Jump on the ❜embed console and click edit on the For You template to customize it in the Feed Builder
-> API first: Get you API key in the >Getting Started page and go to the docs if you want to start playing with the API directly
-> DM @albiverse on Twitter/Farcaster to get added to our Slack channel for devs and ask questions!
-> Follow us on X @mbdtheworld and Farcaster @mbd for updates.