Circuit-AI: A Self-Hosted AI-Agent Language Model Framework for Control Loop Implementation and Simulation

This paper presents an AI-Agent designed to assist with power electronics analysis, simulation, and interface with the hardware to optimize its operation employing LLaMA 3 language model, deployed locally on the NVIDIA Jetson Orin Nano Super. The system employs a self-hosted model to support simulation orchestration, control loop prototyping, and digital twin integration without requiring cloud connectivity. Modular APIs interface with simulation environments and hardware platforms, enabling more streamlined workflows. Engineers can interact with the system via natural language prompts to express design objectives, control strategies, and diagnostic tasks. This approach aims to simplify specific stages of the design process and reduce development overhead. The architecture represents a step toward evaluating the role of generative AI in power electronics workflows under resource-constrained, edge-computing conditions. The proposed system is designed to operate entirely on-device, without internet connectivity (air-gapped), making it especially well-suited for optimizing the performance of defense and industrial systems. Experimental results from the NVIDIA Jetson Orion Super hardware is discussed.