Autonomous Multi-Vendor Research Facility at Scale
A large materials research facility has 150 instruments from dozens of vendors, hundreds of researchers, and a growing network of partner institutions. Every instrument speaks a different format. Every new partner integration is a one-off engineering project. Researchers who want to run an experiment need to understand the specifics of every system they touch — and when a storage backend changes, every file reference across every system breaks.
This is not an unusual situation. It is the default state of most serious research facilities.
Designing the Process
Anyone who understands a step in the research process can build it in Workflow Manager — a researcher, a technician, a domain expert. Workflow Manager supports templates, so a process that runs repeatedly can be defined once and reused. But the point of an autonomous facility isn't rigid repeatability. It's flexibility within a coordinated structure. Workflows can vary between instruments and be tailored to the actual sequence a process requires. The template is a starting point, not a constraint.
Each step in a workflow has a schema — a definition of what information needs to be captured at that stage. Schemas are designed by the subject matter experts who own each step, using their own terminology and structured around their organization's goals. Contextualize does not prescribe what gets captured or what it's called. A metallurgist defines the print parameters the way their team understands them. A characterization specialist defines the inputs for their instrument in the language their lab already uses. The result is a workflow built from real domain knowledge, not a generic template imposed from outside.
Different experts can own different steps. Neither needs to know what the other's instrument does or how it works. Each contributes what they know, and the workflow assembles that knowledge into a coordinated process.
Not every step needs to be automated. Steps that require a hand measurement, a visual inspection, or a human judgment call are marked as human-in-the-loop. The workflow pauses at those points, waits for input, and continues. Automation handles what can be automated; the researcher stays involved where they need to be.
Running Without Knowing How Each Instrument Works
When a researcher initiates a run, the translation between the workflow and each instrument happens through automation configured within Carta. The researcher fills out the form for their step. They do not write run-control files, look up vendor documentation, or manually move data between systems. Each instrument receives the input it expects, in the format it expects — the researcher never touches that layer.
Each instrument integration is configured once. After that, it runs. Adding a new instrument to the facility means building its integration into Carta — and from that point forward, any workflow that includes it behaves the same way.
When results come back, they are immediately associated with the run that produced them. Every result is linked to the request that initiated it, the parameters that defined it, and the steps that came before and after it in the workflow. The data lives in the facility's own storage systems — not Contextualize's. If the facility decides to change their storage backend, one configuration changes in Carta. Nothing else breaks. No file references need to be updated. No downstream systems need to be reconfigured.
Bringing in a New Partner
A partner institution wants to use the facility. Maybe they need access to a piece of equipment their lab doesn't have. Maybe they want to contribute a step in an existing workflow, or build something new on top of capabilities the facility already has.
They do not need to learn the facility's vendor stack. They do not need credentials into the facility's infrastructure. They fill in their steps of the workflow through the same interface every other researcher uses. Carta handles the coordination across the institutional boundary using the same encrypted delivery mechanism it uses internally. The partner's involvement looks, from their side, like filling out a form and receiving results.
Onboarding is fast because the coordination layer is already built. Adding a new partner means adding them to a workflow — not building a new integration from scratch.
Doing the Work That Matters
Before Carta, every new instrument integration was a project. Every new partner was a negotiation over data formats, credentials, and access. Researchers spent time learning systems instead of doing science. When infrastructure changed, things broke in ways that were hard to trace and expensive to fix.
After Carta, the facility runs on a coordination layer that sits above the vendor stack. Instruments are integrated once and speak through Carta. Researchers still work with their instruments — they just stop spending time on the data formats and file management that surround them. Partners slot into existing workflows without a new integration project for every one. The facility's data stays owned by the facility, organized and linked in a way that survives any infrastructure decision they make in the future.
The goal of an autonomous research facility isn't to remove researchers from the process. It's to make sure that when a researcher's expertise is needed, that's what they're spending their time on — not the infrastructure around it.