Engineering knowledge is scattered across drawings, reports, emails and people’s memories. How a generative AI knowledge assistant can make it searchable and useful, and the data, security and accuracy practices that make it trustworthy.
A senior engineer was retiring after thirty years. His knowledge of why certain designs were chosen, which materials had failed in the field and which suppliers could handle difficult parts existed mostly in his head and in decades of scattered files. Younger engineers spent hours searching shared drives for past reports, often unsuccessfully, and sometimes repeated mistakes that had been solved years earlier.
A knowledge assistant, built on generative AI and connected to company documents, can help solve this problem, as a UK design team found.
What a knowledge assistant does
Engineers ask questions in plain language, such as "What draft angle did we use on textured PC/ABS housings?" or "What were the root causes of the warpage problem on the 2021 enclosure?", and the assistant searches relevant documents and answers with references to the sources.
The common approach, retrieval-augmented generation, finds relevant documents first, then uses an AI model to compose an answer from them. It keeps answers grounded in company information rather than general knowledge.
What to include
- Design guides, standards and specifications.
- Design review records, DFM reports and lessons learned.
- Test reports and failure investigations, including 8Ds.
- Material and supplier information.
- Process documentation, work instructions and FMEAs.
Making it trustworthy
- Source references for every answer, so engineers can check the original.
- Clear handling of uncertainty: the assistant should say when it cannot find an answer.
- Good document quality: outdated or contradictory documents produce poor answers. Curate the content.
- Version control: prefer current revisions and flag superseded documents.
- Feedback: let users flag wrong or unhelpful answers.
Security and access
Engineering documents often include confidential customer information. A knowledge assistant must:
- Respect existing access permissions, so people see only documents they are allowed to see.
- Keep data within approved, secure environments.
- Follow the company's AI governance policy, as described in AI governance for engineering firms.
Starting small
- Choose a focused domain, such as plastics design guidelines, material selection or a product family.
- Gather and clean the relevant documents.
- Pilot with a small group of engineers.
- Measure time saved, answer accuracy and user satisfaction.
- Expand to more domains as confidence grows.
Capturing new knowledge
A knowledge assistant is only as good as the knowledge it can access. Encourage teams to record lessons learned, decisions and their reasons in a consistent format. Over time, the assistant becomes a living memory of the engineering organisation.
Not a replacement for engineers
The assistant helps engineers find and use knowledge faster. Judgement, verification and accountability remain with people, as with any AI tool used in engineering work.
