An AI assistant that knows the manuals and the failure history

Challenge

Manuals, procedures and technical documentation were spread across many documents and internal repositories. Finding the right information during a failure took time, and the answers depended on whom the technician asked.

Solution

We designed an AI assistant for internal maintenance staff. It answers questions from internal documents and publicly available technical sources, and cites the source for every answer.

• RAG technology: the assistant first finds the relevant passages in documents and only then composes an answer
• A simple web interface with answer quality ratings directly in the chat
• A strong focus on security and the protection of internal data

Phase 1: Custom chatbot

A custom-built chatbot with a web interface, connected to the GPT-4o mini model. We taught it to work with internal documents in TXT, DOCX and PDF formats.

• Testing directly with the client’s users
• 187 questions answered and 61 answer ratings
• Feedback was used to fine-tune answer quality

Phase 2: Production deployment in MS Copilot

Building on what we learned in testing, we deployed the assistant to production in Microsoft Copilot.

• Connection to the internal SharePoint repository and enterprise systems including SAP
• Data summarisation and report creation
• Additional information from the open web with source references