One post tagged with "LLM"

Scientific datasets are powerful, but exploring them can be tricky and time-consuming. You write boilerplate to list datasets, remember specific syntax, switch mental models between server-side APIs and local operations, and pray you don't accidentally load 50GB into RAM.

That's why we built the Caterva2 Agent — not as a replacement for the analyst, but as a companion that removes the busywork standing between your data and your insight. The core idea is simple: you ask for what you need in natural language, and the agent grounds every answer in real tool calls against the actual data. No hallucinated statistics. No made-up shapes. Every number, every slice, every plot comes from executing real operations on real datasets.

But here's what makes it genuinely useful for scientific work: the agent is designed to sit inside your existing Jupyter notebook exploration workflow, not replace it. You can start from the server, fetch a useful subset, work locally in your notebook with your own transformations, and then continue with the agent from that exact point, since it can see your variables and use them, and you can run your own code on the agent's responses. It's a loop: agent-assisted exploration → manual analysis → agent-assisted exploration, seamlessly switching between server-side operations and local ones as you go.

A short demo of the Caterva2 Agent in action.