AI article reviews assistant

A bot on discord where anyone, with any language, can inquire and discuss about arxiv articles. The bot will further collect information following user requests.

Comment

Inspiration

A friend celebrates their 500+ Scientific AI article reviews. Until now he only shared them as part of his social circles - Facebook, X, LinkedIn, etc. He wanted an inpsiring way to make his articles more accessible - I suggested an assistant that lets you get them.

With AgentCore, we decided to go on an easier UI (Discord), and also open a new science oriented community about AI.

What it does

For now, the bot answers on a dedicated channel. You ask a question and it answers on that channel. You can ask follow up questions.

How we built it

The bot is connected to Bedrock Haiku 4.5 for multilanguage and faster reply, and the aritcle papers on S3, indexed with S3 Vectors

Discord is used as UI.

The code was pushed in layers, so I started with just a discord dummy bot, added a vanilla agent with local bot on Ollama. Then switched to AgentCore, and added features like knowledgeBase. Lastly I added a model on Bedrock instead of Ollama for robustness.

This layered approach, where I also push the prompts I used for each push, is designed for everyone else to copy, either the path, or the result.

Challenges we ran into

Almost none. The most challenging was handling permissions to manage S3.

Accomplishments that we're proud of

The bot is working very well, and people started using it to find existing papers.

What we learned

Working with AgentCore is easy! I know it from the workshop on Amazon that I attended, but the models were also able to use it very well. It means documentation for it is great!

What's next for AI article reviews assistant

Next will be moving the whole agent to AgentCore, to allow other access methods like a site. Also, I started multi-agent information gatherers system for ongoing knowledge growth - I want it to be a multi-agent solution, based on AgentCore.

Lastly, I want the bot to be more educational. Teach concepts and terms through articles and curiousity, and encourage conversation around these topics.

The goal is to serve anyone - without language barrier - for better scientific growth and learning.

Built With

Share this project:

Updates