Recognition and extraction¶
The system reads whatever you give it, in any form: documents, photos, audio, video. This is why you don't have to copy anything by hand.
How it works¶
Do you have your price list only in a PDF scanned years ago? Upload it anyway. The system recognizes the text from the scan, extracts it, and makes it usable for the bot. The same applies to a photo of a menu, a voice message, or a presentation video.
What is recognized, type by type:
| What you provide | What is extracted |
|---|---|
| Digital document (PDF, Word, Excel) | Text and tables |
| Scanned or photographed document | Text, recognized from the image |
| Photos and images | A description of the content, plus any text present |
| Audio (recordings, voice messages) | Transcription of what is said |
| Video | Transcription of the audio |
Where you encounter it¶
- In the knowledge base: everything you upload passes through here before becoming a source.
- In the voice messages customers send to the chatbot: transcribed and understood instantly.
- In callbot calls: the customer's voice becomes text, and summaries are generated from there.
- From the app, when you photograph a document to add it to the sources.
What you need to do¶
Nothing technical: upload the material in the form you have it. Just one tip: the more readable the original is, the better the extraction.
Limits¶
- Handwriting is the worst case: a pen-written note might not be recognized.
- Audio with a lot of noise, overlapping voices, or very technical terms may be transcribed with errors.
- In videos, only the audio is understood: what is seen only on screen, without being spoken, does not become text.
- Very complex tables (merged cells, notes, cross-references) may lose their structure: see Media.
Frequent problems¶
- I uploaded a scan and the bot says imprecise things → open the extract and check: if the recognition made a mistake, correct the text or upload a more readable original. See Verification.
From the app
Photograph a document from the app and the recognition transforms it into a source for the bot.