Extracting BOMs from Electrical Drawings with AI: Azure OpenAI GPT-5.4 + Azure Document Intelligence

2026-04-22 · Source: Microsoft Foundry Blog articles · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Advanced, long

Summary

A multi-stage AI pipeline, combining Azure OpenAI GPT-5.4 and Azure Document Intelligence, has been developed to automate the extraction of Bills of Materials (BOMs) from complex electrical single-line diagrams (SLDs). The pipeline achieves 94.21% accuracy, processing 53 panels across 4 diagram pages in approximately 62 minutes. Key techniques include a "divide and conquer" strategy that processes diagrams panel-by-panel, a two-pass Azure Document Intelligence detection for figure regions supplemented by GPT-5.4 for gaps, and a hybrid GPT-5.4 and Document Intelligence approach for precise text localization. It also features an iterative Locate → Verify correction loop with oscillation detection for panel boundary identification, few-shot visual prompting for spatial concepts, and optimized reasoning levels for different pipeline stages. An alternative SVG vector boundary detection method is also explored for CAD data.

Key takeaway

For AI Engineers developing solutions for structured information extraction from complex technical documents, you should prioritize a multi-stage, hybrid AI architecture. Leverage specialized tools like Azure Document Intelligence for deterministic tasks and precise geometry, while reserving powerful but more expensive models like Azure OpenAI GPT-5.4 for semantic understanding and iterative refinement. This approach optimizes both accuracy and cost, enabling robust solutions for challenging visual parsing problems.

Key insights

Decomposing complex visual tasks into smaller, hybrid AI stages significantly improves accuracy and efficiency.

Principles

• Decompose problems aggressively for tractability.
• Optimize inputs before scaling compute.
• Hybrid AI solutions outperform pure LLM approaches.

Method

A multi-stage pipeline uses Azure Document Intelligence for initial detection and precise text, supplemented by Azure OpenAI GPT-5.4 for semantic understanding, iterative verification, and few-shot visual prompting.

In practice

• Use Azure Document Intelligence for fast, cheap detection.
• Employ few-shot visual prompts for domain-specific visual concepts.
• Implement iterative verification with oscillation guards.

Topics

• Electrical Single-Line Diagrams
• Bill of Materials Extraction
• Azure OpenAI GPT-5.4
• Azure Document Intelligence
• Multi-stage AI Pipeline

Code references

• jihys/electrical-sld-bom-extraction

Best for: AI Engineer, Machine Learning Engineer, MLOps Engineer

Related on AIssential

• Closing the Last Mile in Document AI: Improving Extraction Quality in Azure Content Understanding — Labeled examples significantly boost document field extraction accuracy in generative AI systems like Azure Content Understanding.
• The PDF Paradox: Why Document Parsing Is Still Hard — And Why the Hybrid Stack Is Winning — PDF's fixed-layout design, while excellent for rendering, creates a significant parsing challenge for AI.
• Benchmarking Open-Source Layout Detection Models for Data Snapshot Extraction from Institutional Documents — Generic layout models fail to extract meaningful data snapshots from complex institutional documents.
• From 4 Weeks to 45 Minutes: Designing a Document Extraction System for 4,700+ PDFs — A hybrid AI and rule-based system efficiently extracts data from complex PDFs, balancing cost, speed, and accuracy.
• Forget GraphRAG: A 4B AI does the work NOW — Agent-native knowledge orchestration via K1's 4B LLM and multimodal graph overcomes GraphRAG limitations for scientific reasoning.
• Granite 4.0 3B Vision: Compact Multimodal Intelligence for Enterprise Documents — Granite 4.0 3B Vision offers compact, modular multimodal intelligence for enterprise document understanding.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Microsoft Foundry Blog articles.