Skip to main content

Best practices

Adam - Menta's AI mentor - draws on your documents to ground his responses and feedback. The clearer and better‑organised your documents are, the better that grounding will be. None of the following is required, but each tip will noticeably improve the responses and feedback your learners get.

Prepare clean, readable documents​

  • Upload the real document, not a scan or photo of one. A text document the system can read directly works far better than an image of a page. If you only have a scanned PDF, run it through optical character recognition (OCR) first so the text is selectable.
  • Make sure the file opens normally before you upload it. As covered in Uploading files, anything password‑protected, encrypted, or corrupted cannot be used.
  • Prefer text‑based formats. Information conveyed only in images, charts, or diagrams may not be picked up. If a diagram carries something important, add a short text caption or summary near it.

The system works best when each piece of content is self‑explanatory in isolation, because it retrieves passages rather than whole documents.

  • Use clear headings and subheadings. They help the system understand how a document is organised and keep related information together.
  • Keep each section focused on one topic. Short, well‑labelled sections retrieve far better than long, mixed ones.
  • Spell out context instead of relying on "it" or "the above." A paragraph that stands on its own - naming the product, policy, or process it refers to - is easier to surface accurately.
  • Use plain, consistent terminology. If your organisation has a preferred term for something, use it consistently so searches line up with your wording.
  • Expand acronyms at least once. Write the full term alongside the abbreviation so both can be found.

Organise your collections thoughtfully​

How you split your documents across knowledge collections has a big effect on how relevant Adam's answers are. A few patterns work well:

  • Group related material together. Put documents that belong to the same topic or purpose in the same knowledge collection - for example, one collection for product documentation and another for your sales methodology. This keeps searches focused and relevant.
  • Keep one language per collection. A knowledge collection has a single language, and the files in it should be written in that language. If you have material in several languages, create a separate collection for each - see Match the file language to the collection.
  • Keep collections current. Remove or replace outdated documents so Adam isn't drawing on superseded information.

Avoid duplicate content​

When the same document appears in two knowledge collections that are both used by one course, that course sees the content twice. Adam can then treat one source as if it were two independent ones, making an answer seem more certain than the evidence really supports.

tip

Keep each document in a single knowledge collection where you can. If you need the same material in several places, prefer sharing one collection across courses over copying the file into multiple collections. Within a single collection, exact duplicates are detected and skipped automatically - but content split across different collections is up to you to keep tidy.

The same caution applies across languages: avoid assigning the same document in more than one language to a single course. The two versions say the same thing, so the same information can be retrieved twice - crowding out other useful context while adding little of its own. Keep one language per collection, and assign each course the collection that matches the language it's taught in.