I recently launched a daily newsletter that curates the top content of the day of a decentralised social network called Farcaster.

Farcaster is a decentralized social media platform that aims to revolutionize the way people interact and share information online, think like Twitter but without Elon or anyone to block your account.

A few weeks ago, I invited some crypto/web3 friends to Farcaster, and they struggled to find the value. Common responses were: "I don't get it," "It's too niche," or "too much shill." I knew there was good quality content on Farcaster, so I started manually curating the top content per day and sending it by email. Overnight I started to get hundreds of subscriptions.

However, manual curation became painful. Going through trending topics, channels, etc., every morning was not scalable. So, I started querying the casts, filtering, and using AI to select the best content for myself. After a few iterations, I've achieved a high-quality automated classifying system using AI and built an email system that sends the emails after my last supervision.

Unfortunately there is no email platform that allows this type of automation, so I have to build the entire platform from scratch. Content curation, user subscription, email sending and more.

Tech stack: - Vercel for db and netxjs hosting - Resend for email delivery - Zeplo for queueing emails - LLMs for curation and spam detection: gtp-4-turbo and claude-3-sonnet

I was not expecting this to grow that fast and I was hoping to do it manually for quite some time, but I guess it is a nice problem to have.

I tried before to create other newsletters but failed to reach content market fit. I guess my take away from this experiment is that nowadays to grow at this rate you have to go first very niche, second you probably need a lot of customisation.