RAG search across 20 years of documentation

Semantic search over manuals, drawings and service reports. Quotation drafting dropped from days to hours.

A mid-sized machine builder with around 400 employees needed answers faster than its archive could provide them: 20 years of manuals, drawings and service reports spread across drives and systems.

Starting point: Quotation and service teams spent hours hunting for previous projects, specifications and known failure patterns. The knowledge lived in a few experienced heads, and every quote began with searching instead of writing.

Approach: A RAG pipeline over the entire document base: preparation and chunking of the sources, semantic search with embeddings, answers with source citations instead of bare text generation, and access rights taken from the existing directory service. Quality was evaluated against a gold set of real enquiries; guardrails, monitoring and cost control made the system production-ready.

Outcome: Quotation drafting dropped from days to hours. Service enquiries get answers with evidence instead of gut feeling. The system runs in production, operated and evolved by the client’s own team.

• Industry: machine engineering
• Duration: 4 months to production
• Services: AI/GenAI Implementation, AI Adoption Coaching