replyguy
Your 24/7 reply guy.
The problem replyguy solves
People waste hours of their life in scrolling through the internet in search of content that is not relevant to them.
Why does this problem exist?
Lack of personalisation and context in the content that is being consumed.
What are the possible solutions to this problem?
The generic solution to this is to ask user to define their interests and preferences, and then based on that, the content is personalized.
But this is not a good or scalable solution. No one has time to define their interests or select the option to "Not Interested".
How does ReplyGuy solve this problem?
ReplyGuy uses AI intelligence to understand each and every user's post and its intent, and if the post is reply-worthy, then the agent will fetch the most relevant content across Farcaster and reply to the user's post with a link to it.
How it works
Single click to subscribe to the ReplyGuy in the Farcaster miniapp.
That's it! The user does not need to provide any information or context.
Now, cast like before and get the most relevant content in your feed.
Tech Stack
- Neynar APIs for interacting with the Farcaster Protocol.
- Eigen Layer AVS on Base and Ethereum.
- Gemini and OpenAI for generating embeddings and replies.
- Supabase for storing the data.
Contracts
The following AVS contracts have been deployed on Base Sepolia (L2) and Ethereum Holesky (L1):
NOTE
We have deployed AVS on the Base Sepolia and Ethereum Sepolia testnet, due to the cost and setup of the reward distribution mechanism for validators, as we are not generating revenue right now.
But we have a solid plan to generate revenue that is to provide a service to subscribers to boost their posts' visibility on Farcaster to a targeted audience. This is a well-requested feature and is present on every social media platform at a very high price, but we will be able to provide it at a very low price with much better results.
Challenges we ran into
In starting, we were happy with the responses from the AI and in getting our first 120 Subscribers.
Link: https://warpcast.com/yourreplyguy/0xa9cfb518
But when we looked for the feedback from the users, we found two major issues:
- The reply frequency was really high.
- Sometimes the reply was not relevant to the post.
So we sat, screamed through the night and came up with the following solution:
Conditional reply based on the post.
Examples:
- If the post is just a reply to some other post or any empty quote post, then don't reply.
- If the post is in a channel, prioritise the content discovery from the channel.
Another issue was verifying that the data is being fetched from the trusted servers and replies are being generated by the AI and doesn't have any human intervention.
To fix this, we came up with the following solution:
Verifiable Agent using Eigen Layer.
We built an Eigen Layer AVS on Base and Ethereum to verify the data and the replies posted by the agent, using Reclaim Protocol and Othentic stack.
Tracks Applied (3)
AI
Mini-apps
Consumer
Technologies used
Cheer Project
Cheering for a project means supporting a project you like with as little as 0.0025 ETH. Right now, you can Cheer using ETH on Arbitrum, Optimism and Base.