EngiAI: Multi-Agent Framework for LLM-Driven Engineering Design
A new benchmark suite and multi-agent system called EngiAI has been introduced for evaluating LLM-driven engineering design. The benchmark comprises three dimensions: a workflow benchmark with seven prompt styles targeting cognitive demands like direct tool use and semantic disambiguation; a Retrieval-Augmented Generation (RAG) benchmark with gated scoring for parameter selection; and an HPC benchmark for end-to-end ML training orchestration on a SLURM cluster. EngiAI is a reference implementation built on LangGraph, coordinating seven specialized agents through a supervisor. The framework addresses the lack of evaluation standards for multi-agent systems combining simulation, retrieval, and manufacturing preparation.
Key facts
- EngiAI is a multi-agent system for LLM-driven engineering design.
- The benchmark suite has three dimensions: workflow, RAG, and HPC.
- Workflow benchmark includes seven prompt styles.
- RAG benchmark uses gated scoring for retrieval contributions.
- HPC benchmark evaluates ML training orchestration on SLURM.
- EngiAI is built on LangGraph.
- Seven specialized agents are coordinated by a supervisor.
- The framework targets simulation, retrieval, and manufacturing preparation.
Entities
Institutions
- arXiv