An assistant that knows PEMA's 28 years of work and speeds up preliminary quotes

The challenge

PEMA has designed, delivered and installed fencing systems since 1998. Over 28 years it built up a body of work from hundreds of projects: offers, designs, a way of running client conversations. The problem wasn't a lack of knowledge, but getting to it. On every new quote or unusual inquiry, someone had to search the documents by hand for what the company had done before and how.

That cost twice. Once, when an experienced person broke off their own work to answer a new employee a question asked for the tenth time. And again, when a preliminary quote dragged on, because someone had to reconstruct a similar, previously completed project from scattered files.

• 28 years of work scattered across documents, hard to use quickly.
• Preliminary quotes delayed by the manual search for similar projects.
• Repetitive questions from new hires pulled the team out of current work.

The solution

We built an internal assistant based on company knowledge. Under the hood it's RAG: the assistant doesn't guess, it answers from PEMA's real documents and doesn't make things up. You ask in natural language, with no folder digging. Step by step it looks like this.

1. We gathered the company's knowledge in one place: PEMA's offer, projects and way of communicating.
2. The documents went into a vector database, so the assistant searches by meaning, not by exact word.
3. An employee's question runs into a workflow in n8n that searches the knowledge base and assembles an answer from the company's documents.
4. We tuned the language and answer style to PEMA, so it sounds like the company, not a generic model.
5. A convenient interface handles the whole thing, so everyone uses it, not just the technical department.