Enhancing LLMs for Manufacturing Information Extraction

Automating manufacturing workflows requires accurately extracting structured information from technical specifications, a task critical for compliance and safety. While previous studies have explored applying Large Language Models (LLMs) to manufacturing for tasks such as knowledge sharing and document management, they have not addressed the specific demands of structured specification extraction required for automation. We introduce ManuExtract, a benchmark dataset for extracting item–property–value triplets in manufacturing, and ManuLLaMA, a fine-tuned LLaMA-3.1-8B model trained on this dataset. Our approach combines LLM-generated annotations refined by domain experts with data augmentation to address data scarcity. ManuLLaMA shows strong cross-lingual performance, advancing automation and reproducible benchmarking. Our code and datasets are available at https://github.com/MLAI-Yonsei/ManuExtract .