Normalisation
Structured extraction from PDF and CAD, with version and model as metadata.
Work
Industrial
From technical documentation to operational support
Thousands of manuals, diagrams and service bulletins held answers that field technicians could not find in time.
Applied AI
RAG
Field service
- Role
- AI system design
- Market
- International
- Duration
- 9 months
- Team
- 5 people
01 — The challenge
The knowledge existed. It was simply unreachable.
Documents in different formats and languages, overlapping versions, no reliable way to know which revision applied to a specific machine.
The non-negotiable requirement was traceability: every answer had to cite its exact source.
02 — Approach
Verifiable retrieval
Answers always anchored to citable passages, never free-floating.
Field use
An interface built for tablets, partial offline and busy hands.
03 — Outcome
−71%
average time to find a procedure
18k
technical documents indexed
94%
answers accepted by technicians
Technology
AzureLLM orchestrationOCRPythonEvaluationVector store