NVIDIA Holoscan is a platform for building real-time AI applications at the edge, spanning fields like medical imaging and robotics. The ALCHEMI Toolkit, introduced earlier this year, has streamlined the development of simulation workflows by leveraging AI coding agents. These agents can generate code and simulation pipelines using examples, documentation, and tools typically used by engineers. This approach reduces the time and effort required to create complex simulation environments.

The ALCHEMI Toolkit is designed to work with Holoscan’s companion repository, HoloHub, which contains a collection of reference applications and components. This repository demonstrates the capabilities of the platform and provides developers with a foundation for building their own AI applications. The toolkit supports a variety of tasks, including materials simulation, by integrating with existing simulation stacks and providing accessible interfaces for researchers and developers.

The agentic processing time for building simulation workflows was recorded at 40 minutes, a significant reduction compared to traditional methods. The toolkit also draws on lessons from 45 generated pipelines to provide practical guidance on building trustworthy GPU-accelerated simulation workflows. These pipelines are compatible with CUDA 13 and require Python 3.11, PyTorch 2.8, and Linux as the primary operating system.

The use of AI coding agents in simulation workflows introduces considerations around cost, vendor lock-in, and governance. While the toolkit reduces development time, it also depends on specific hardware and software configurations, such as CUDA 13 and Linux. This can lead to increased dependency on NVIDIA’s ecosystem. Additionally, the integration of AI agents into the development process raises questions about the reliability and transparency of the generated code, which must be carefully managed to ensure trustworthiness.

The ALCHEMI Toolkit and Holoscan are still in the early stages of development, with ongoing improvements and updates expected. The system has shown promise in accelerating simulation workflows, but it remains a work in progress. As the toolkit evolves, it will be important to address challenges related to compatibility, scalability, and the long-term sustainability of AI-generated code in scientific and engineering applications.