Workshop: XLOOP 2026: The 8th Annual Workshop on Extreme-Scale Experiment-in-the-Loop Computing
Authors: Venkateswaran Shekar, Yevgen Matviychuk, Eliot Gann, and Daniel Allan (Brookhaven National Laboratory)
Abstract: Scientific user facilities such as the National Synchrotron Light Source II (NSLS-II) need reliable paths from beamline acquisition to machine-learning analysis and scientist-facing interaction. We present a Tiled-centered architecture that publishes scientific data and inference results through one interface while keeping compute, model-registry, and language-model services replaceable. As an institutional partner in the American Science Cloud (AmSC), we developed Emblase to demonstrate an application that uses AmSC-provided compute, model-hosting, and LLM inference services in a live NSLS-II Soft Matter Interfaces (SMI) scattering workflow. Data acquisition creates Bluesky runs that Tiled exposes to an event-driven inference service. The service resolves a versioned physics-informed machine-learning model, processes incoming frames, and writes embeddings and associated products to Tiled for dashboard inspection. Authenticated Tiled tools support grounded natural-language interaction. The architecture keeps live experimental state behind one scientific interface while allowing deployment-specific choices for compute and AI services.
Back to XLOOP 2026: The 8th Annual Workshop on Extreme-Scale Experiment-in-the-Loop Computing Archive Listing
Back to Full Workshop Archive Listing