Shanghai Jiao Tong University

Solution

Shanghai Jiao Tong University and Baidu AI Cloud built an AI for Science platform that combines generative AI, scientific data management and trusted data sharing in one research environment. ERNIE provides the base model layer, while the BaiGe Heterogeneous Compute Platform optimizes training and inference across the university's research workloads. On top of that, the data middle platform and AI middle platform manage scientific datasets, model development and deployment workflows, and the privacy-computing platform enables trusted sharing and federated modeling across disciplines. The resulting architecture underpins a scientific-data open platform that can answer molecular queries, retrieve reaction processes and literature links, and support forward reaction prediction, knowledge Q&A and model evaluation. Rather than leaving AI expertise siloed inside one lab, the platform industrializes compute, data governance and model services so researchers in chemistry, law and urban science can run repeatable AI-assisted experiments on a shared foundation.

Data flow

Scientific datasets and model artifacts are governed through the data and AI middle platforms, scheduled on BaiGe compute resources for training and inference, then served to researchers through AI-for-Science tools that return reaction knowledge, predictions, evaluations and multi-domain model outputs.