Skip to main content

Chatbot for Dialogue with Technical Manuals

Machinery manufacturer (anonymized)Machine Learning & AIIndustry

Chatbot for Dialogue with Technical Manuals

Challenge

As a partner to a provider of technical information, we took on the challenge of building an AI chatbot for a major player in the machinery manufacturing industry. The goal was to enable fast and efficient information retrieval in extensive technical manuals for service technicians.

Project setup

The scope of the project was limited to a couple of service manuals — around three manuals, together at least 3,000 pages. Limiting the number of documents kept the project focused, while the page count was enough for the problem to be real: finding the right passage in 3,000 pages by hand is exactly the work the solution was meant to replace.

Solution

Our solution was based on RAG technology (Retrieval-Augmented Generation), which combines a search engine with an AI chatbot. By breaking down the manual's content into smaller segments and indexing these in a vector database, we were able to create an effective method for finding and presenting relevant information.

Every answer comes with a source reference to the right section of the manual. The precision of those references was the entire value of the solution — a service technician who cannot verify an answer against the manual page will not use it.

The solution was built on Azure, with the manual content indexed in a vector database in the same environment.

Outcome

Our work resulted in a functional chatbot that could handle specific requests from an extensive manual. In the pilot, we saw clear effects for service technicians and support teams:

  • Response time for common manual questions dropped from minutes to seconds.
  • First-line support could resolve more cases directly without escalation.
  • Service technicians reached the right manual sections faster, reducing downtime during troubleshooting.

After the PoC the client took over development and carried the solution forward in-house. That was the point of the setup: prove that RAG worked against their actual manuals, at a contained cost, and hand over something they could build on themselves.

Learnings

The greatest insight from this project was the importance of high-quality data. The success of an AI-driven solution heavily relies on the quality of the underlying data. Close collaboration with the client is crucial to ensure access to and processing of relevant data.

The solution shows how RAG can move document search from minutes to seconds in industrial environments — and how an AI assistant can become a working tool for service technicians rather than just a demo.

Getting the chance to work with such new 'cutting-edge' technology was fantastic. We are very pleased with the outcome and look forward to continuing to develop the solution.

Porträtt av Jonathan Köre
Jonathan Köre
Head of Machine Learning

Technology

  • Machine Learning
  • Large Language Models
  • Data Processing
  • AI Agents
  • Vector Databases
  • Azure

Next steps

Want to use RAG/LLMs to make manuals and knowledge searchable in your organization? Contact us and we’ll set up an intro call.

More case studies

AI chatbot for geothermal permit applications

We built an AI chatbot for Herrljunga municipality that guides citizens through geothermal permit applications via chat, interactive maps, and automa...

Read more

SaaS platform for board work and meeting management

We helped Candoris Nova build a SaaS platform for board work. The platform unifies agendas, decision logs, ownership tracking and AI support – from p...

Read more

⚡ We answer within 24h!

Tell us what you need to solve

By submitting this form, you accept our privacy policy.