In manufacturing companies, two isolated information worlds emerge systematically along the order process: structured data in the ERP system on one side, technical and qualitative content in certificates, inspection reports, and delivery notes on the other. Modern AI systems for automated information processing can close this gap.
How this works in practice and what role multimodal large language models play in this is described in a specialist article for IT&Production. The article was published online on 7 May 2026 and appears in print edition 4/2026. The article is published in German.
The Most Important Insights
- Multimodal LLMs enable semantic understanding even with unknown document layouts for the first time, significantly reducing the need for predefined data fields.
- Classic text recognition (OCR) takes on the role of grounding — it links extracted content to the original document, thereby preventing misinterpretations.
- At order entry, AI-assisted systems automatically detect technical requirements such as required inspection certificates or standard specifications and attach these in structured form to the order.
- At goods receipt, inspection values can be validated not only against standards but also against order requirements and supplier histories — gradual quality drifts become visible at an early stage.
- Before goods dispatch, the linked database enables automatic completeness checks of documentation as well as drastically simplified traceability in the event of complaints.
- The software solution from Helm & Nagel uses a containerized infrastructure (Docker/Kubernetes), can be connected to existing ERP systems via REST API, and can be operated as SaaS or on-premise.
- Validation and extraction rules can be formulated in natural language — this reduces dependency on IT specialists and sustainably anchors domain expertise in the production environment.
About IT&Production
IT&Production is the leading German-language trade magazine for industrial IT and Industry 4.0 in the processing industry. It is published ten times a year by TeDo Verlag GmbH in Marburg and is aimed at production and plant managers, IT managers, and domain experts from automation, maintenance, and process optimization.