Automated Data Readiness for Scientific AI
06:00 · July 7, 2026 · arXiv cs.AI RSS

Leadership computing facilities steward large-scale scientific datasets that routinely require substantial transformation before serving as AI training data. However, no existing framework fully unifies automated transformation, readiness assessment, provenance tracking, and agent-native deployment. We present REDI, an open-source framework that addresses this gap through a unified five-stage pipeline (ingest, preprocess, transform, structure, and output) with per-stage instrumentation for reproducibility and deployment as an agent-callable skill; companion tool SetGo automates FAIR compliance and catalog publication. Evaluated across climate, proteomics, materials science, and nuclear fusion, REDI transforms all datasets from raw to AI-ready, with outputs validated against domain-expert references, and preliminary results show near-ideal parallel scaling to 100 nodes on Frontier for the climate case. Provenance-instrumented profiling reveals file I/O as the dominant pipeline cost, with format selection a first-order optimization lever. These results establish REDI as a cross-domain platform providing automated data readiness for scientific AI, transforming data preparation bottlenecks into reproducible, reusable community assets.
Summary
Leadership computing facilities routinely manage petabyte-scale scientific datasets that must undergo extensive cleaning, validation, and restructuring before they can serve as training material for AI models. The Readiness Engine for Data Integration (REDI) is an open-source framework that automates this transformation through a single, instrumented five-stage pipeline—ingest, preprocess, transform, structure, and output—while recording provenance at every step. The pipeline can be invoked directly by agentic coding environments, allowing reproducible execution without manual scripting.
A companion utility, SetGo, complements REDI by enforcing metadata standards that satisfy the FAIR principles and by publishing the resulting catalogs to repositories such as CKAN or Hugging Face. Together the two tools close the gap between conventional data stewardship and the stricter structural and semantic requirements of large-scale model training.
When applied to datasets from climate science, proteomics, materials modeling, and nuclear fusion, REDI produced AI-ready outputs that matched domain-expert references. On the Frontier supercomputer the climate workload exhibited near-linear scaling to 100 nodes. Detailed profiling showed that file I/O accounted for the largest fraction of runtime, making format choice—Zarr, NPZ, or ADIOS—a primary lever for further performance gains. By converting ad-hoc preparation steps into documented, reusable assets, REDI reduces duplicated effort and improves the auditability of scientific AI workflows.
Why it matters
This research is highly relevant to Dutch AI researchers and institutions because it provides an open-source, scalable solution for scientific data preparation while explicitly automating FAIR compliance—a critical standard in the European and Dutch research ecosystems.







