Democratizing climate data science: How Columbia University’s LEAP center built AutoClimDS on AWS

This article describes AutoClimDS, an agentic AI system developed by Columbia University's LEAP center with AWS to democratize climate data science research.

• AutoClimDS enables researchers to conduct climate data science using natural language without coding expertise
• Multi-agent architecture with specialized agents for data discovery, acquisition, analysis, and verification
• Built on knowledge graph integrating climate datasets from NASA and other sources
• Uses Amazon Bedrock, Neptune, Lambda, S3, and Textract for cloud infrastructure
• Fine-tuned ClimateBERT classifier achieves 99.17% accuracy linking observational data to climate models
• Designed for reproducibility, modularity, and scalability of climate research workflows
• Lowers technical barriers enabling broader participation in climate research

AutoClimDS demonstrates how AI agents can democratize access to complex scientific data, enabling researchers to focus on hypothesis generation rather than data wrangling.