AI consultant in Eindhoven

The ERP system holds orders, actual costs per production batch and routings: the sequence of operations for each product. A MES, the system that records what happens at each machine, logs downtime and faults. There are also machine logs, quality-control measurement reports and non-conformance reports.

These sources often exist but are not connected. The first task is to find the link: which order number belongs to which machine run?

It uses retrieval-augmented generation (RAG). The system splits documents into sections, finds the passages relevant to each question and has a language model base its answer on them. If you make the system cite the document and paragraph for every answer, you can check each one.

Anthony Chat works this way for Fokker Services Group. You can read about distinguishing between similar contracts in our blog on internal LLMs and RAG.

At Fokker Services Group, maintenance was mainly scheduled at fixed intervals and based on engineers’ experience. This sometimes meant maintenance happened too early or too late. A planning tool now combines maintenance history, sensor data and operational data to produce a prediction.

The lesson for manufacturing: engineers’ experience does not disappear; it becomes an input. Their knowledge became features, the characteristics the model uses in its calculations. Read the maintenance planning case study.

Yes, and the data is often already available. For Fokker Services Group, a model predicts what an invoice amount should be based on the work delivered, then compares that prediction with the actual amount. Large discrepancies appear in a dashboard, so staff only need to check the exceptions.

The same approach could work for subcontractors’ invoices or actual costs per production batch. Read the invoice-checking case study.

AI-hub Brainport has a workstream dedicated to the digital transformation of SMEs using AI. TU/e’s EAISI institute researches AI in machines, such as robots and medical equipment. OEMs, suppliers and knowledge institutions come together at Brainport Industries Campus in Eindhoven.

Their focus is on research, knowledge sharing and regional collaboration. We build an application for your own process and manage it if you want us to. The two approaches work well together.

Because experienced people are in short supply. In 2024, the Dutch government reported that the microchip sector in Brainport Eindhoven faced a shortage of staff, among other challenges. Through Project Beethoven, the national and regional governments are therefore investing, among other things, in training more technicians. Meanwhile, much of the knowledge about machines, customers and exceptions resides with experienced engineers.

An assistant built around work instructions, or a model that draws on their maintenance experience, helps keep that knowledge usable when someone leaves or changes roles.

If you produce few variants and your schedule rarely changes, your ERP planning module is often enough. A fixed maintenance interval is fine if faults are rare and cause little damage. And if the answers are in a handful of documents, a good folder structure will help more than an assistant.

A dashboard using ERP and MES data often provides enough visibility into downtime and backlogs. See our Power BI dashboards.