Representative engagement
An assistant for technicians. It combines machine manuals, maintenance logs and past interventions to suggest likely causes and procedures. It drafts the work order afterwards.

- Client
- A Swiss precision manufacturer
- Sector
- Industry & Manufacturing
- Duration
- 10 weeks
- Services
- Agentic Systems · Adoption & Enablement
01Challenge
Unplanned downtime on critical machining centres was costly, and diagnosis depended on a few senior technicians. Maintenance logs were written in shorthand, in French and German, and manuals ran to thousands of pages.
Newer technicians took longer to diagnose faults and repeat failures were common. Plant data could not leave the site network.
02Approach
Forward-deployed engineers spent time on the shop floor with technicians to understand how faults are actually diagnosed. We normalised ten years of maintenance logs and linked them to machines, components and manual sections.
The assistant runs on site. Technicians describe a symptom in their own words on a tablet. The assistant returns likely causes ranked by past occurrence, with the procedure and parts. It drafts the work order when the job is done.
03The system
An on-premises assistant over manuals, maintenance history and parts data, accessible from the shop floor, with safety-relevant procedures referenced to the official manual.
Components
- 01On-premises model deployment on the plant network
- 02Normalised maintenance history linked to machines and components
- 03Manual and procedure retrieval with page references
- 04Symptom-to-cause ranking from past interventions
- 05Work order drafting into the maintenance system
- 06Technician training and feedback loop
Stack
- Open-weight LLM on premises
- Maintenance system integration
- Hybrid search
- Tablet web app
04Results
A Swiss precision manufacturer
- −35%
- Mean time to diagnose on targeted machines
- −20%
- Repeat failures within 30 days
- 100%
- Data kept on the plant network
Representative engagement. Client withheld under NDA. Details are anonymised and combined so that no client can be identified; metrics are those measured on the engagement.
Related work
Next step
Tell us which process takes your team the most time. In a thirty-minute call, we can tell you if the sprint is a good start.


