SIParCS 2026 - Anna Scribner

Using LLM Chat to Simplify User Interactions with GDEX Data

Recorded Talk

Scientific datasets contain valuable information that can help answer important research questions. However, accessing and visualizing these datasets often requires programming skills and familiarity with specialized data formats, making them difficult for many people to use. This project explores how large language models can help make NCAR's Geoscience Data Exchange (GDEX) datasets more accessible through a natural-language interface. Users describe what they want to explore in natural language, for example, "What has the temperature in Boulder been over the past ten years?" The application interprets the request, generates a plotting script, runs it in a safe environment, and returns both the visualization and the generated script. While this work is still experimental, its goal is to reduce the technical barriers to exploring scientific data and make these datasets more accessible to students, researchers, and anyone interested in learning from them.

Mentors: Nathan Hook, Harsha Hampapura, Jason Cunning, Eric Nienhouse, Chia-Wei Hsu

Slides and poster