ExpandingRG

Inspiration

Research groups that write their own simulation code lose knowledge constantly. How to build the code, why a setting was changed, which bug was already fixed: it lives in people's heads and leaves when they graduate. New members spend weeks catching up, and the same problems get solved twice.

What it does

ExpandRG gives each researcher a private AI agent and the group a shared knowledge base, the Super Brain.

• Your agent helps you understand the code, set up and run simulations, and diagnose problems.
• Your work stays private until you approve a finding, then it's shared with the group.
• New members get auto-generated lessons that connect papers to the code, with equations and diagrams.
• Runs are monitored for problems and compared on interactive graphs.

How we built it

A web dashboard and a VS Code extension, backed by a server that keeps private and shared data apart. A small local service reads only the folders a researcher allows, so code stays on their computer. CREAO powers the AI agents: one per researcher, plus the group's Super Brain.

Challenges we ran into

• Keeping AI responses fast despite a slow hosted agent API.
• Real research code is messy: unusual build files, unlabeled outputs.
• Making sure AI explanations are accurate, by checking them against the code and the papers.
• Balancing sharing with privacy.

Accomplishments that we're proud of

• It works on our lab's real research code, not a toy example.
• Sharing is always the researcher's choice.
• One-click install, with over 600 automated tests.

What we learned

In research the bottleneck is knowledge, not compute. AI help is only useful when it's verified and when people trust where their data goes.

What's next for ExpandRG

Support for computing clusters, smarter onboarding, and piloting it with more research groups.

Built With

• creao
• crossref-api
• fastapi
• javascript
• katex
• node.js
• plotly
• pydantic
• pypdf
• pytest
• python
• react
• sqlite
• typescript
• uvicorn
• vite
• vscode-extension