Images, PDFs with photos, catalogs

Photo catalogs, laid-out menus, flyers: materials made to be looked at, not read by a machine. You can still use them as bot sources.

How it works

You upload the PDF catalog: 40 pages of product photos with names, descriptions, and features. The system extracts the text, reads the tables, and describes the images. From then on, if a customer asks "do you have the blue fabric armchair?", the bot can answer using the catalog information.

The bot can also show images in the conversation when the channel allows it: the product photo along with the answer.

What you need to do

1. Upload the file to the knowledge base, just like any other document.
2. After indexing, verify with a few questions about the catalog content.
3. If an important part is only inside a photo (a sign, a label), check that it was read correctly.

Limits

• Very complex tables (merged cells, marginal notes) may lose some parts: it is better to rewrite critical information as plain text, see Writing content.
• A grainy photo or a very low-quality PDF results in poor extraction.
• Prices in catalogs get outdated: if they change often, external data read in real time is better.

Frequent problems

• The bot cannot find a product that is in the catalog → it is probably described only by the photo: add a couple of lines of text about that product.
• The bot shows the wrong image → report the conversation: see The bot made a mistake.

From the app

From the app you can photograph a flyer or a catalog page and add it to the sources.