LLM4EDGE: AGENTIC AI FOR EFFICENT ENGINEERING

To increase the efficiency of engineering processes with large language model (LLM) agents that act as AI assistants and agentic co-engineers. Agents can use models running on local edge hardware, in the cloud, or on enterprise-hosted infrastructure, or combine these approaches depending on data confidentiality, connectivity, latency, cost, and energy constraints, to improve:

• Offline: operation, maintenance and process optimization of industrial plants in highly constrained environments without connectivity, where data must not leave the site
• Online: development processes in large engineering organizations (e.g. semiconductor, industrial automation, robotics and automotive companies), from requirements, design and simulation (including EDA for electronics and chip design) to testing, validation and documentation

The main challenges and topics of investigation are:

• Hybrid local/cloud operation: choosing the right model per task based on connectivity, data sensitivity, latency, cost and energy
• Agentic engineering workflows: planning, orchestration and verification of multi-step tasks with human-in-the-loop control
• Standardized and safe tool integration via the Model Context Protocol (MCP) to connect engineering and plant tools such as simulation, CAD/PLM, EDA, requirements and test management.
• Efficient models for edge hardware: model selection, quantization, distillation and domain adaptation under compute, memory and energy limits
• Agent harness: safe execution environment, guardrails, traceability and systematic evaluation for industrial use
• Measuring efficiency: methods and KPIs to quantify gains in lead time, manual effort and quality
• Trustworthiness: hallucination control, explainability, IP protection and regulatory compliance (e.g. EU AI Act)