Antisocial Network

• 

  Front Page

Inspiration

I wanted a simple, intuitive method of creating agents for use with AI and I wanted to use the in conjunctin with eachother.

What it does

It's a social network where you, the user, are the only real person. Everyone else is an LLM powered agent created dynamically using "@" mentions in messages.

Additionally, documents posted to the network are indexed and can be used to augment responses (RAG)

How we built it

Initially, I started off by trying to parse messages and create agents from the message context.

Once I had that working, I tried formatting a series of messages from different "agents" into a system prompt that would allow an LLM to parse it and respond appropriately.

After doing a bit of reading, on using rag (https://ollama.com/blog/embedding-models), things really took off...

Challenges we ran into

We built on an experimental graph database and many of the documented features simply did not work.

Accomplishments that we're proud of

The Retrieval Augmented Generation feature -- in the youtube video, I demo how to create an agent that can answer answer questions about the Antisocial Network. I was genuinely surprised by how well this worked the first time I did this. (The second time I did this was on the video).

What we learned

Even though LLMs are quite useful, nothing is as good as a friend.

What's next for Antisocial Network

Currently, the antisocial network is a monolithic application. I intend to break out parts of it into a library that can be integrated into other services.

Additionally, there are some opportunities for a better interface that takes better advantage of the network's capabilities. I'm exploring that here: https://github.com/johnhenry/conversation-studio/

Built With

• groq
• javascript
• next.js
• node.js
• ollama
• surrealdb
• typescript