Help desk. Product documentation available when it is needed

01. Starting point

The client sells devices that someone has to install, commission, and service. Every product has its own document set: user manual, installation guide, technical documentation, catalog sheets. Documents exist in three language versions and are updated with each product revision.

The audience for this knowledge is not the end customer, but the installer and the trade partner. That difference matters, because an installer needs a specific piece of information at a specific moment, usually standing next to the device, not browsing a catalog at a desk in the evening.

The earlier portal ran on WordPress and amounted to a product list with downloadable files attached. There was no search across document content, no clear versioning, no assistant, and no training sign-ups. A user who did not know the exact product name had to scroll the list.

That was not an execution mistake; it was a consequence of the tool. WordPress is a system for publishing content and does that job well. A technical documentation portal is a different class of application: it needs file-content indexing, revision control, parallel language versions tied to one product, and accounts created on the manufacturer’s side. Building that on a CMS means a chain of plugins, each owning a slice of the problem, none understanding the rest, with constant updates and compatibility risk. At this scale it is faster and cheaper to write an application than to maintain a workaround.

02. Solution

We built a portal where all product documentation lives in one place, always in the current version, available after login.

Accounts are created by an administrator on the manufacturer’s side, so access is limited to partners and installers, not casual visitors. Every product has its own card collecting all related files, ordered by document type, version, and language.

On top of that sit two search mechanisms. Search looks not only at file names, but also at their content, including PDF contents. An AI assistant answers questions asked in plain language, grounding the answer in the documentation of a specific product.

The portal also handles sign-ups for product trainings, so a partner has knowledge and a way to deepen it in one place.

03. How it works

The product card as the reference point

Every product has one card with all related documents. Versioning is visible directly: the user knows which revision they are downloading and whether a newer one exists. When the manufacturer publishes an update, nobody needs a separate notice, because the partner pulls the document from the source, not from a year-old copy.

All materials are maintained in parallel in Polish, English, and German. Changing language does not dump the user back to the home page; it shows the same document in the chosen version when available.

Search across document content

Search indexes file contents, not only metadata. A query for a connector marking, error code, or parameter name finds the document even when the phrase appears only on page twenty of the manual. Results cover both products and individual documents, with suggestions for typos and inflection.

An assistant that answers from the documentation

The assistant takes a question the way an installer would ask technical support over the phone. In the reply it points to the right manual or describes directly what to do in the situation, based on installation documents.

The source boundary is critical. The assistant answers only from the manufacturer’s documentation, not from the model’s general knowledge, and points to the document the answer came from. The installer can verify it before acting at the device. For a question outside the docs, the assistant signals that instead of guessing.

Trainings

Partners sign up for product trainings directly from the portal, in the same place they use the documentation.

Usage statistics

The portal records which files are downloaded, how often and by whom, and which products draw the most interest. The manufacturer also sees what the assistant is asked most often.

Those data go beyond reporting. A file downloaded repeatedly by the same installer usually means the information inside is hard to find. A repeated question to the assistant points to a documentation gap. Portal usage becomes feedback on the quality of the materials themselves, and along the way on which products need sales support versus training support.

Technologies used
  • Application: Next.js, TypeScript, React
  • Database: PostgreSQL
  • Search: Meilisearch, indexing of PDF file contents
  • Assistant: language model with semantic search over documentation, answers limited to manufacturer materials
  • Access: accounts created by an administrator, session authentication
  • Deployment: Docker, nginx

04. Results

Current documentation stopped being something you have to ask for. A partner finds it alone, at any hour, without support as a middleman and without the risk of working from an outdated revision.

Questions whose answers sit in the manual stopped landing with people. Technical support handles unusual cases, not redirects to a chapter.

Three language versions take the same effort as one. Publishing an update is uploading a file to the product card.

The manufacturer stopped working blind. They know which documents installers open, which products generate the most questions, and where documentation needs improvement. The earlier portal delivered none of that.

See also

AI Social Media manager
AI Social Media manager. Months of posts planned ahead
AI application
AI application. Campaign results from five sources in one dashboard
Product configurator
Product configurator. Shelter quote and visualization in one tool