2026/01/09 14:07:24
AI Revolutionizes Scientific Research at Berkeley lab’s Advanced Light Source
In the rolling hills of Berkeley, California, a groundbreaking request of artificial intelligence is dramatically enhancing the efficiency and capabilities of the Advanced Light Source (ALS), a world-renowned particle accelerator. Researchers at Lawrence Berkeley National Laboratory have successfully deployed the “Accelerator Assistant,” a large language model (LLM)-driven system designed to streamline operations and accelerate scientific finding. This innovation isn’t confined to Berkeley; it’s part of a broader Department of Energy (DOE) initiative, Genesys, aimed at replicating this success across other U.S. particle accelerator facilities.
The Challenge: Complexity and Downtime at a World-Class Facility
The ALS is a complex machine. It accelerates electrons to near the speed of light, generating intense beams of ultraviolet and X-ray light used for over 1,700 scientific experiments annually. These experiments span diverse fields, including materials science, biology, chemistry, physics, and environmental science [[1]]. However, maintaining optimal operation is a meaningful challenge. The ALS control system manages over 230,000 process variables, and even minor disruptions can halt experiments and lead to costly downtime. As Thorsten Hellert, a staff scientist at Berkeley Lab, explains, “It’s really important for such a machine to be up, and when we go down, there are 40 beamlines that do X-ray experiments, and they are waiting.”
Introducing the Accelerator Assistant: An AI-Powered Solution
The Accelerator assistant addresses this challenge by leveraging the power of large language models. Powered by an NVIDIA H100 GPU and utilizing CUDA for accelerated inference, the system taps into the collective knowledge of the ALS support team.It can access and process information through various LLMs,including Gemini,Claude,and ChatGPT.Crucially, the Accelerator Assistant doesn’t just provide information; it can actively solve problems, wriet Python code, and even autonomously prepare and run experiments – all wiht the option of human oversight.
How it works: Context Engineering and Secure Integration
The system’s effectiveness stems from a sophisticated approach to “context engineering.” operators interact with the Accelerator Assistant via a command line interface or Open WebUI, allowing access from control room stations and remotely. the underlying framework, Osprey, developed at Berkeley Lab, ensures safe and agent-based AI implementation within the complex control system. Each user is authenticated, and the system maintains personalized context and memory across sessions.
the accelerator Assistant connects to a vast database of process variables, a ancient data archive, and Jupyter Notebook-based execution environments. This allows it to understand the current state of the accelerator and draw upon past experiences. Inference can occur locally, using Ollama on an H100 GPU, or externally through a secure gateway (CBorg) connecting to models like ChatGPT.This hybrid approach balances speed, security, and access to cutting-edge AI capabilities.
A key element is the integration with EPICS (Experimental Physics and Industrial Control System), a widely used distributed control system in scientific facilities. this integration allows the AI to interact directly with the accelerator hardware while adhering to established safety constraints. Essentially, the system translates conversational input into clear, actionable tasks, leveraging external knowledge and personalized user data to provide relevant and accurate responses.
Beyond ALS: Expanding the Impact of AI in Scientific infrastructure
The success of the Accelerator Assistant has spurred its adoption beyond the ALS. As part of the DOE’s Genesys mission, the framework is being deployed at other U.S.particle accelerator facilities. Moreover, Hellert is collaborating with engineers at ITER, the world’s largest fusion reactor in France, to implement the framework for fusion energy research. A collaboration with the Extremely Large Telescope (ELT) in northern Chile is also underway, demonstrating the broad applicability of this technology.
Real-World Impact: Accelerating Scientific Discovery
The benefits of the Accelerator Assistant are already becoming apparent. The research team has demonstrated that the system can autonomously prepare and execute a multistage physics experiment, reducing setup time and effort by a factor of 100 [[1]]. This translates to more efficient use of valuable beam time and faster scientific progress.
Benefiting Humanity Through Scientific Advancement
The ALS, empowered by the Accelerator Assistant, is contributing to breakthroughs with far-reaching implications.During the COVID-19 pandemic, ALS research played a crucial role in characterizing a neutralizing antibody against SARS-CoV-2, aiding in the development of effective therapeutics. Furthermore, ALS science has contributed to advancements in climate research, especially in the development of metal-organic frameworks (MOFs) for carbon capture – work that was recognized with the 2025 Nobel Prize in Chemistry. In planetary science, ALS measurements of asteroid Bennu have provided insights into the origins of water and life on Earth.
The Future of Accelerator Science: Autonomous Operation with Human Oversight
Looking ahead, Hellert envisions a future where the Accelerator Assistant can autonomously run the ALS, with human operators providing oversight and approval. The creation of a comprehensive wiki documenting all operational processes will be crucial in enabling this level of automation. “On these high-stakes scientific experiments, even if it’s just a TEM microscope or something that might cost $1 million, a human in the loop can be very critically important,” Hellert emphasizes.
The Accelerator Assistant represents a significant step forward in the application of AI to complex scientific infrastructure.By automating routine tasks, accelerating problem-solving, and enabling new levels of efficiency, this technology is poised to unlock new discoveries and address some of the world’s most pressing challenges.
