From Auto Research to AI

Ajay started working for Mercedes Benz Research and Development India, which not only gave him first-hand experience practicing his craft, but also exposed him to German work culture and offered him the chance to start learning the German language. He parlayed that experience into a graduate program in Germany, earning his master’s degree in computational sciences and engineering at the Technische Universität Braunschweig (TU Braunschweig). 

As he got deeper into the field, Ajay realized that the skill set he had been developing was broadly applicable within the greater HPC ecosystem. “In a way, I realized that the, core principles of research and development are what I would call domain agnostic,” he said. “To me, that means that once you start to think about what you are actually doing with a code base, you realize that many research domains are all trying to do the same thing: you want to improve the efficiency of codes so they will go faster or use less energy without sacrificing the quality of your simulations,” he said. 

As Ajay grew his HPC knowledge, the field was also in the process of major change. Specifically, artificial intelligence (AI) – a technology that had long been developed and incubated at public HPC centers – was coming into its own, and the rise of generative AI promised to radically alter how researchers could use large-scale computing resources to accelerate their research. Ajay was introduced to neural networks during his graduate studies and saw the power in using these methods to further accelerate computational research. Neural networks operate similarly to how the human nervous system processes information and form the foundation for large-scale AI model development. “This was my second major eureka moment, and I realized that if you program neural networks well, they can make accurate predictions based on what you teach it,” he said. “We were free to choose our own research subjects, and I decided that I wanted to focus more on machine learning as the core of my graduate studies.”

Working as a user support specialist at LRZ, Ajay’s work focuses on the intersection of his passions, but he also gets to support research teams at an institution focused on scientific outcomes. “One of the best parts of working at LRZ is that we are funded by public and for research projects,” he said. “As a result, we are not interested in researchers’ money, or trying to monetize researchers’ data – we are only providing a service that is otherwise difficult for many research institutions to build or maintain on their own.” In his role, Ajay sees supporting users in the AI era as a two-pronged issue: LRZ staff must help researchers with less experience using HPC or other large-scale computing resources learn to use these systems efficiently while also supporting long-time HPC users in adapting their workflows to take advantage of AI. He pointed out that many research projects rely on legacy codes that have been collaboratively developed over years. Accordingly, he sees a core part of his role as helping users develop AI models that can be integrated into these long-running research tools.

Since LRZ is one of the three computing centers that comprise the Gauss Centre for Supercomputing (GCS), Ajay also benefits from a larger network of computational science experts for knowledge sharing, collective problem solving, and sharing best practices. Together with other user support specialists at the High-Performance Computing Center Stuttgart and Jülich Supercomputing Centre, he feels motivated and well-positioned to deliver on the core GCS mission – supporting researchers at all levels in solving the pressing scientific and engineering challenges of today and tomorrow. “One of my key responsibilities is providing access to these valuable resources,” he said. “Even if a person is running a relatively simple algorithm but has a bigger idea they know cannot run on a laptop or smaller compute cluster, we want to be able to help them get started. I want to foster a close loop between researchers and our team, so that we’re not just waiting for the user to come to us with a problem, but we are investing the time in helping develop solutions that can proactively benefit teams and, in some cases, the larger research community.” (Eric Gedenk | GCS)