An AI-native, multi-agent framework that autonomously completes power electronics design tasks through natural-language prompting.
Agentic AI Framework
Customized for power electronics design with multi-agent workflow
Autonomous Design
From natural language prompts to engineering-ready deliverables
Paradigm Shift
Transforming power electronics design with AI-native workflows
What is PE-GPT
An agentic AI framework customized for power electronics design.
Capabilities
From natural-language prompts to engineering-ready deliverables, PE-GPT is powered by an agentic architecture designed for power electronics.
The architecture integrates agentic reasoning, retrieval-augmented knowledge, model zoo–driven design, simulation verification, and fine-tuning loops.
Agentic Task Orchestration
Decomposes high-level requests into coordinated design, verification, and reporting steps.
Knowledge-Grounded Reasoning (RAG)
Grounds design decisions in retrieved technical knowledge, documentation, and domain data.
Model Zoo–Driven Design
Leverages analytical, numerical, and data-driven models for parameter synthesis and trade-off analysis.
Simulation-Based Verification
Validates design outputs through physics-based simulations to reduce unrealistic or invalid results.
Physics-Informed Learning
Incorporates physical constraints and domain structure to improve robustness and interpretability.
Data Loop & Fine-Tuning
Continuously adapts agents using design data, waveforms, and task-specific feedback.
Design Cases
PE-GPT has been developed and validated through multiple representative power electronics design cases, including converters, magnetics, and device-level characterization.
Education
PE-GPT supports AI-native power electronics education through open resources, project-based learning, and engineering-ready deliverables.
In the midst of the rapid advancement of artificial intelligence, it is time to prepare the next generation of the power electronics workforce. Future power electronics engineers and researchers will be distinguished not only by their domain expertise, but also by their ability to work with AI as an engineering collaborator.
As an agentic framework, PE-GPT provides educational resources freely through its open-source release on GitHub. It enables the training of students to develop an interdisciplinary mindset and to effectively harness AI capabilities in solving real-world power electronics engineering problems.
Through educational efforts in collaboration with the University of Arkansas Power Group, the PE-GPT framework has been deployed in a graduate-level course on AI for power electronics design. Within a 16-week semester, postgraduate students with zero prior AI expertise independently built, extended, and demonstrated functional PE-GPT-based solutions, emphasizing task decomposition, verification, and engineering-ready deliverables.
Student Project Tracks
Student teams adapted and extended the PE-GPT framework into the following directions:
Student Contributors
The following students contributed to the development and demonstration of PE-GPT-based educational projects:
Samuel Fakunle, Brooke Scott, Hamood Ur Rehman;
Alec Broomfield, Sumina Neupane, Xianfeng Jiang;
Md Majharul Islam, Weiping Fu, Eric Allee;
Tory Sutton, Fasih Uddin, Xianglai Zeng;
Daixin Chen, David Setor Agogo-Mawule, Wesley Schwartz.
Partnerships
We collaborate with partners across industry, academia, and education to advance AI-native power electronics design from research to real-world impact.
Industry Collaboration
We work with industry partners to explore AI-native and agentic design workflows for practical power electronics applications, including autonomous converter design, magnetics optimization, semiconductor characterization, and engineering productivity tools.
Academic & Research Collaboration
We welcome collaborations with academic and research institutions interested in advancing the foundations of agentic and physics-informed AI for power electronics, including benchmarking, methodological research, and joint publications.
Educational Collaboration
Building on successful classroom deployment, we collaborate with universities and instructors to integrate PE-GPT into graduate courses, laboratories, capstone projects, and interdisciplinary AI–power electronics curricula.
Partnership inquiries are handled through the Get in Touch section.
Resources & Citation
Key publications, code resources, and recognitions related to PE-GPT.
How to Cite
F. Lin et al., "PE-GPT: A New Paradigm for Power Electronics Design,"
in IEEE Transactions on Industrial Electronics,
vol. 72, no. 4, pp. 3778–3791, April 2025,
doi: 10.1109/TIE.2024.3454408.
PE-GPT was recognized as a deployed application demonstrating real-world impact of AI-native, agentic systems for power electronics design.
Get in Touch
For partnership, research, or educational collaboration inquiries, please leave your information below. We will follow up accordingly.
How we respond
- Your inquiry is reviewed and routed based on the selected interest.
- We may follow up for clarification or additional details.
- A short discussion may be scheduled to align next steps.