A product catalogue AI assistant lets a sales or technical team ask a question in plain language and get the answer from the manufacturer’s own catalogues, with the drawing on screen. We built one for the local office of a European aluminium building-systems manufacturer. It answers from about 1,300 pages of window, door and curtain-wall catalogues plus the price lists, understands article numbers, shows the relevant technical drawing next to the answer and puts the price in the same reply.
The problem: the answer is on page 247, in Polish
The local team supports fabricators, installers and architects. Their questions are specific: which gasket fits this profile, can these two profiles be joined, what is the net price of this article, which accessories come with this system. The answers exist, but they sit in system catalogues of several hundred pages each, written mainly in Polish with English in places, and in separate price lists.
Much of the knowledge is not in the text at all. It is in the drawings: profile cross-sections, assembly details, dimension tables. A keyword search cannot read a drawing, so finding an answer means knowing the catalogue well or leafing through it. New colleagues need time to learn where things are, and experienced ones spend their time on lookups instead of on customers.
What we built
An assistant that reads the catalogues the way a technical person does, text and drawings together:
- Every page was read once, including the drawings. Around 1,290 catalogue pages were rendered as images, and each page got a written description of what the drawing shows, so drawings become searchable.
- Catalogues and price lists form one knowledge base. Over 3,300 indexed passages: catalogue content for two major product systems and the standard and additional price lists.
- It recognises article numbers. Codes are found even when typed with spaces or partial, and an exact article match takes priority over a general search.
- It answers in the right order. A price question gets the price table first, then catalogue details. A technical question gets the technical answer and drawings first, with the price at the end.
- It shows the source. The pages that contain the article are shown inline, never unrelated pages, and three follow-up questions are suggested.
The answer is written by a language model that looks at the retrieved text and the page images together, so it can explain what a drawing shows, not only quote the text around it.
What changes for the team
| Before | With the catalogue assistant | |
|---|---|---|
| Finding a technical detail | Leafing through several hundred pages | A question in plain language |
| Drawings | Found by knowing the catalogue | Shown next to the answer |
| Price and technical data | Two separate documents | One reply, in the right order |
| Language of the source | Mostly Polish | Answers in the language of the question |
| New colleagues | Need to know where things are | Can ask without knowing the catalogue |
The team still owns the relationship with the customer. What changes is the time between the question and a confident answer.
Where people stay in control
The assistant always shows where the answer comes from, so a person can check the drawing or the price line before passing it on. It answers from the documents it was given: when a catalogue is updated, the new version is added and the knowledge base is refreshed. Access is protected by a login.
An honest limit: a price list is only as current as the last file loaded. The assistant shows prices from the version it has and does not replace the official quotation process.
Where else this works
Any company whose know-how lives in long technical documents can use the same approach:
- Building products and HVAC distributors: product selection and compatibility questions from supplier catalogues.
- Industrial equipment: maintenance manuals and spare-part lists in one searchable assistant.
- Construction contractors: specifications and standards answered with the page reference.
Related use cases: Automated technical quotation · Insurance product documentation AI · Insurance policy document extraction. All Internal assistants & tools use cases · How we deliver this: AI applications
FAQ
Can it read technical drawings, not just text?
Yes. Every catalogue page was processed as an image, and the answer is written from both the text and the page images, so questions about cross-sections and assembly details can be answered.
What happens when a user types an article number?
The assistant detects the code, normalises it if it was typed with spaces, and looks for an exact match in the price lists and catalogues before running a general search.
Where does the data stay?
The catalogues and price lists are loaded into a private knowledge base behind a login. The documents are not used to train any model.
Want your own catalogues to answer questions with the drawing on screen? Talk to us