Document Training

Steps to train documents:

Go to "Files" section in Content Tab.
Click on choose File to upload single/multiple files.
Click on "Save & Train".

Note:

Allowed file format are .pdf, .txt, .csv.
Maximum file size can be 20 MB.
Maximum 20 files can be uploaded in a single training.

Advanced Pdf Parser

Advanced mode introduced for parsing pdfs containing images and complex tabular data to enhance content training capabilities. Advanced parsing is done using GPT-4O model. User can select Advanced option in the file upload modal which appears when user clicks on "Choose File" in Content Tab.

Value Delivered:

Increased run time AI accuracy in response generation for PDF-based content
Reduced manual effort during content training as no pre processing required for complex pdf
Supports real-world business documents like, brochures, reports and more

Basic vs Advanced Usage Guidelines

Basic: Default Parsing option for all the pdf files. Simple pdfs which contains some images and mainly text can be easily parsed using Basic option.

Advanced: Use Advanced option only when PDF contains lots of images and nested tabular data (High training time & cost as compared to Basic)

Key Limitations:

Parsing accuracy will be based on clarity of the image text
Slow & Costly content extraction
Final content extraction accuracy depends on the performance of LLM Model (Gpt-4O)
Extraction is subjected to Azure content moderation policy - if it detects certain words in the content then it can get blocked
Model may not perform optimally when handling images with text of non-Latin alphabets, such as Japanese or Korea
The model may misinterpret rotated or upside-down text and images
Not suggested for medical use cases like interpreting CT scan images