Speaker
Description
To support Phase II of CSNS and overcome key limitations of existing data management system (DMS), a new generation DMS tailored for AI4Science has been designed. A new metadata standard for photon and radiation experiments will be applied to enable unified management across the entire experiment lifecycle and multiple facilities. The system federates NoSQL and vector databases, supporting both keyword-based and semantic-based metadata search. Tiered storage employs a peer-to-peer mechanism to manage data transfer at the file base between tape and disk, with on-demand compression during file movement and storage. AI agents leveraging locally deployed large language models (LLMs) are under development: one extracts and completes metadata from diverse data sources, another generates automated experiment analysis reports, and a third, an experiment planner, uses retrieval-augmented generation (RAG) and simulation to assist in experiment design. A new data portal will integrate these capabilities, offering users a unified interface for experiment planning, simulation, data search, visualization, analysis, and retrieval.
