Trust, Trace, Transform: Data-Centric Materials Discovery with FPGA Inference, Kubernetes Orchestration, and Dataerai Curation
Significant advances in AI-assisted materials design have largely accrued in near-equilibrium regimes. Many technologically relevant systems-functional oxides, correlated phases, metastable alloys, and defect-engineered architectures-require non-equilibrium synthesis in which transient kinetics, spatial gradients, and defect dynamics dominate. We present a data-centric architecture for real-time control of non-equilibrium synthesis that integrates curation, physics-informed learning, and actuation across instrument, edge, and cloud resources.
The approach comprises: (i) automated scientific data curation that elevates heterogeneous telemetry (images, spectra, events, control logs) into provenance-rich, schema-validated datasets; (ii) low-latency edge inference via quantized, FPGA-deployable surrogate models enabling sub-millisecond feedback; and (iii) synchronized multimodal pipelines that fuse high-rate diagnostics (e.g., >500 Hz RHEED and plume imaging) to stabilize experimental processes within tight latency budgets. We demonstrate real-time analysis and closed-loop control in scanning-probe spectroscopy, steering of pulsed-laser-deposition plume dynamics, and microsecond-scale actuations relevant to plasma/tokamak control, each requiring algorithm-hardware co-design.
To render these capabilities routine, we introduce Dataerai, a provenance-centric curation and orchestration layer that converts raw instrument streams into AI-ready, auditable corpora, addressing the foundational bottleneck in AI-driven materials discovery: reliable access to curated data. Dataerai provides immutable lineage, fine-grained access control, and retention/compliance policies, while supporting governed data sharing for cross-lab insights. Taken together, this curate-compute-control paradigm establishes a portable, reproducible platform for beyond-equilibrium synthesis in which curated, trustworthy data and explicit reward functions-rather than ever-larger interpolative models-govern agentic scientific AI that can transform the rate of discovery.
Bio: Joshua C. Agar is the founder and CTO of Dataerai (pronounced data-ray), a platform that enables compliant curation, governed sharing, and high-fidelity search of AI-ready scientific data. He previously served as an assistant professor at Drexel University and Lehigh University. Trained as an experimental materials scientist, Agar’s research advances the integration of AI algorithms, computing infrastructure, and cyber-physical systems to achieve real-time, closed-loop control in materials synthesis and microscopy. His methods have been deployed across particle and plasma physics, materials science, and fluid dynamics, emphasizing scalable, provenance-rich data pipelines and edge-to-cloud inference. He maintains domain expertise in the synthesis and characterization of ferroelectric thin films, linking non-equilibrium processing with data-centric, reproducible discovery.