Synthetic materials are everywhere, from the chips in our smartphones to the desks in our offices. To design or discover a new material for a particular application, materials engineers often create dozens or hundreds of samples and analyze their properties at the microscopic level.
Unfortunately, time on instruments like electron microscopes is expensive, the gathered data are stored locally, and the sophisticated data analysis necessary to effectively guide the next steps often requires a supercomputer.
“Data transfer to cloud storage or supercomputers is usually slow, taking time away from instrument usage,” said Department of Materials Science and Engineering (MSE) Professor Gerd Duscher. “If we networked an instrument with supercomputers from the start, researchers could begin sophisticated data analysis right away—and it would lay the foundation for AI-assisted microscopy.”
Duscher and Weston Fulton Professor Sergei Kalinin have been working to automate microscopy research for several years. Their joint PhD student, Utkarsh Pratiush, applies his background in programming and hardware integration to build human-AI collaborative instruments.
“Microscopes have traditionally been treated as individual tools operated by individual experts,” said Kalinin. “As microscopy becomes increasingly controlled by machine learning workflows, optimization algorithms, and AI agents, the instruments themselves need to become connected.”
Pratiush has spearheaded development on Asyncroscopy, an open source package that provides remote data access and AI orchestration for electron microscopes. In December 2025, Pratiush began collaborating with researchers at Brookhaven National Laboratory (BNL) to integrate Asyncroscopy into Tiled, BNL’s widely used networking tool for scientific instruments.
“Anybody can go on GitHub right now to download and run Asyncroscopy,” Pratiush said. “I’m now working with an industry collaborator to learn the kind of scenarios they are dealing with, so we can all solve this problem (together).”
Shared Problems, Shared Solution
At first, Pratiush tried to network microscopes and develop their AI capabilities on his own. However, software development takes a lot of trial and error. He became frustrated at his slow progress—and the knowledge that a tool developed by just one person can’t address the wide range of problems the scientific community encounters.
He was delighted to learn about Tiled, BNL’s remote data access network for synchrotrons (particle accelerators), which addresses many of the same issues he was trying to solve.
Just as taking higher-quality photos takes up more memory on a digital camera, high-quality microscope and synchrotron files take up a tremendous amount of space. Remote access to the files is not useful to scientists if the files themselves can’t be stored on personal or lab devices.
Tiled solves this problem by letting scientists preview the data and download single frames, storing only the most important or relevant images to their own computers.
The tool also lets scientists search images by their metadata, or background information about the state of the instrument when the picture was taken. This lets researchers confirm that images taken on different instruments, or months apart, will be taken with the same settings.
When Pratiush, Kalinin, and Duscher reached out to the BNL team to ask about collaborating on a microscope expansion to Tiled, they were given an enthusiastic welcome.
“Joining the community where they’re building this kind of software is much more efficient than doing it alone. It’s also easier to find domain experts to come in and contribute,” Pratiush said. “We went back and forth with the BNL team exploring the project and iterating on it. It has been really fun working together.”
AI-Powered Imaging Labs
Working with the support of BNL researchers, Pratiush developed electron microscopy data adapters and made upstream software improvements to Tiled, resulting in the release of Asyncroscopy. Having demonstrated that the package provides network-based access to microscopy datasets, Pratiush is now working to create a unified ‘language’ that can be used on multiple types of microscopes that give insight into different material properties.
Kalinin, Duscher, and Pratiush’s long-term goal is to integrate different types of microscopes at different locations into one orchestrated network. While one AI-assisted, self-driving microscope can save a lot of time and create valuable insights, a self-driving lab would compound those benefits.
“Federated microscope networks would allow workflows, analysis tools, operational experience, and downstream analytics to be shared across laboratories,” Kalinin said. “These instruments could learn from one another and ‘work’ with humans to enable better experimental decisions and accelerate discovery in real time.”
Contact
Izzie Gall ([email protected])