Frequently Asked Questions

This page is not accurate! Each answer needs to be validated.

General Questions

What is Docling for IBM watsonx?

Docling for IBM watsonx is a fully managed document intelligence service that converts complex documents (PDFs, images, Office files) into AI-ready formats like Markdown, JSON, and HTML. It's built on the open-source Docling toolkit and provides enterprise-grade infrastructure for production workloads.

How is this different from the open-source Docling?

Docling for IBM watsonx is a managed service that eliminates the need to deploy and maintain infrastructure. Key differences:

Managed Infrastructure - No need to set up servers, manage dependencies, or handle scaling
Enterprise Support - Dedicated support channels and SLAs
API-First - RESTful API and Python SDK for easy integration
Automatic Updates - Always running the latest models and improvements
IBM Cloud Integration - Native integration with watsonx.ai, Watson Discovery, and other IBM services

The open-source Docling is ideal for local development and experimentation, while Docling for IBM watsonx is designed for production deployments.

What file formats are supported?

Docling for IBM watsonx supports a wide range of document formats:

PDF documents - Including scanned PDFs with OCR
Images - PNG, JPEG, TIFF, BMP, GIF
Microsoft Office - DOCX, PPTX, XLSX
HTML - Web pages and HTML documents
Markdown - MD files
Text files - TXT, CSV

For the most up-to-date list of supported formats, check the API documentation.

What are the file size limits?

Maximum file size: 50 MB per document
Batch processing: Up to 100 documents per request
Page limits: No hard limit, but processing time increases with document complexity

For larger files or special requirements, contact support.

API & Integration

Do I need to poll for results?

When using the REST API directly, yes - you submit a conversion request, receive a task_id, then poll the status endpoint until completion.

However, the Python SDK handles this automatically. It polls in the background and returns the final result, making it much simpler for most use cases.

How long does conversion take?

Typical conversion times:

Simple PDFs (text-based, < 10 pages): 2-5 seconds
Complex PDFs (tables, images, 10-50 pages): 5-15 seconds
Large documents (50+ pages): 15-60 seconds
Scanned documents (OCR required): 10-30 seconds per page

Processing time depends on document complexity, page count, and current system load.

What happens if conversion fails?

Failed conversions return a status of failure with an error_message field explaining the issue. Common reasons:

Unsupported file format
Corrupted or invalid file
File size exceeds limits
Timeout due to extreme complexity

The error message will guide you on how to resolve the issue.

Can I process documents in parallel?

Yes! You can submit multiple conversion requests simultaneously. The service handles concurrent requests and queues them appropriately. The Python SDK makes this easy with standard Python concurrency patterns.

How do I handle rate limits?

The service implements rate limiting to ensure fair usage:

Rate limit: 100 requests per minute per API key
Concurrent requests: Up to 10 simultaneous conversions

If you hit rate limits, implement exponential backoff in your retry logic. The Python SDK includes built-in retry handling.

Output & Quality

Which output format should I use?

Choose based on your use case:

Markdown - Best for RAG applications, human readability, and LLM consumption
JSON - Best for structured data extraction, programmatic processing, and preserving document metadata
HTML - Best for web display and maintaining visual formatting

You can request multiple formats in a single conversion.

Does it preserve document structure?

Yes! Docling for IBM watsonx preserves:

Layout - Reading order, columns, sections
Tables - Structure, merged cells, headers
Lists - Nested lists, numbering, bullets
Formulas - Mathematical equations (LaTeX format)
Images - Embedded or referenced
Metadata - Document properties, page numbers

This structural preservation is critical for high-quality RAG and information retrieval.

How accurate is table extraction?

Docling uses specialized deep learning models for table recognition and achieves high accuracy on complex tables including:

Merged cells and spanning rows/columns
Nested tables
Tables with irregular structures
Tables in scanned documents

For best results, ensure source documents have clear table boundaries and readable text.

Can it handle scanned documents?

Yes! Docling automatically detects scanned content and applies OCR when needed. For best OCR results:

Use high-resolution scans (300 DPI or higher)
Ensure good contrast and lighting
Avoid skewed or rotated pages
Use supported image formats (PNG, JPEG, TIFF)

What languages are supported?

Docling supports 100+ languages for text extraction and OCR, including:

All major European languages
Chinese (Simplified and Traditional)
Japanese, Korean
Arabic, Hebrew
And many more

Language detection is automatic - no configuration needed.

Performance & Optimization

How can I optimize for speed?

For faster processing:

Use the low-latency option - Set "low_latency": true in options for real-time applications
Process smaller documents - Break large documents into sections if possible
Use appropriate output formats - Markdown is typically faster than JSON
Batch similar documents - Process documents of similar type together
Cache results - Store converted documents to avoid re-processing

What is low-latency mode?

Low-latency mode ("low_latency": true) optimizes for speed over accuracy:

Faster processing (typically 2-3x faster)
Suitable for real-time agentic workflows
May have slightly reduced accuracy on complex layouts
Best for text-heavy documents without complex tables

Use standard mode for maximum accuracy on complex documents.

Can I process documents from S3?

Direct S3 integration is coming soon. Currently, you can:

Download from S3 to your application
Submit to Docling for IBM watsonx via API
Store results back to S3

The upcoming batch mode will support direct S3-to-S3 processing for large-scale ingestion.

Security & Compliance

How is my data secured?

Encryption in transit - All API calls use HTTPS/TLS 1.3
Encryption at rest - Documents and results are encrypted in storage
Temporary storage - Documents are deleted after processing (configurable retention)
API key authentication - Secure token-based authentication
Network isolation - Service runs in isolated IBM Cloud VPC

How long are documents stored?

Processing - Documents are stored temporarily during conversion
Results - Available for 24 hours after completion (configurable)
Automatic cleanup - All data is automatically deleted after retention period

For compliance requirements, contact support about custom retention policies.

Yes, Docling for IBM watsonx is GDPR compliant. IBM Cloud infrastructure meets GDPR requirements, and you maintain control over your data. For specific compliance questions, consult IBM's compliance documentation or contact support.