A consortium of 18 research, academic, pharmaceutical and philanthropic organizations is building open-source AI tools to help dementia scientists analyze data that rarely reaches public models. C-BRAIN’s system combines centralized computing with federated learning, which lets models learn from institutional data without moving the underlying records outside their owners’ systems.
The group has produced two tools. OpenScientist takes a research question and dataset, proposes hypotheses, runs analyses and generates an audit trail. InsightEngine compares hypotheses with published literature, user-provided data and a biomedical knowledge graph, then links claims to evidence so researchers can inspect them.
C-BRAIN says published case studies found that OpenScientist completed some analyses in minutes for less than $10 per run, while domain scientists estimated the same work would take weeks or months. Those comparisons come from the project and do not establish that the tools can independently produce valid discoveries. Fast analysis still needs expert review, reproducible methods and protection against errors in source data.
The consortium currently uses 64 on-premises Nvidia data-center GPUs alongside Microsoft Azure capacity. It plans to combine neuroimaging, molecular measurements, biomarkers and longitudinal clinical records in a broader model. The immediate value is a shared, auditable research workflow—not a demonstrated treatment or cure for dementia.