Reproduce “GraphRAG for Engineering Diagrams: ChatP&ID Enables LLM Interaction with P&IDs” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Achmad Anggawirya Alimin, Artur M. Schweidtmann
ChatP&ID converts DEXPI smart P&IDs into Neo4j knowledge graphs and retrieves graph context for LLM answers. On one single-page diagram, GPT-5-mini with ContextRAG scored 0.91 on a four-dimension judge rubric, versus 0.88 with direct Proteus-file context. Mean task cost and elapsed time were $0.0044 and 24.33 seconds, versus $0.0342 and 52.05 seconds. The rubric is not human-verified accuracy; results are limited to this diagram.
Key insights
- Representing P&IDs as structured knowledge graphs consistently improves LLM performance and dramatically reduces computational costs compared to direct image or file ingestion.
- GraphRAG tools provide a more accurate and cost-effective approach for P&ID interaction, significantly reducing hallucinations compared to multimodal or raw file inputs.
- The LLM-as-a-Judge evaluation method offers more reliable and interpretable accuracy scores for factual graph queries than traditional semantic similarity measures.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.