Parse Document

​

Overview

The Parse Document endpoint processes PDF, DOCX, and PPTX documents using various chunking strategies to break them into meaningful sections. This endpoint is designed for document analysis, RAG applications, and content organization workflows.

​

Request

​

Response

​

Chunking Strategies

​

Semantic Chunking (Default)

Uses advanced AI to intelligently group related content together. Best for maintaining context and meaning.

​

Fixed Size Chunking

Splits text into fixed-size chunks with configurable overlap. Useful for consistent chunk sizes.

​

Page-Based Chunking

Splits documents by pages (PDF only). For other formats, falls back to paragraph chunking.

​

Paragraph Chunking

Splits text based on paragraph boundaries. Good for maintaining natural text flow.

​

Heading Chunking

Splits text based on detected headings and section structure. Ideal for structured documents.

​

zChunk Semantic Chunking

Advanced semantic chunking using LLaMA logprobs for boundary detection. Provides high-quality semantic boundaries.

​

Hybrid Chunking

Multi-modal processing with table extraction, image analysis, and enhanced metadata. Most comprehensive option.

curl -X 'POST' \
  'https://visionapi.unsiloed.ai/chunking' \
  -H 'accept: application/json' \
  -H 'X-API-Key: your-api-key' \
  -H 'Content-Type: multipart/form-data' \
  -F 'document_file=@document.pdf;type=application/pdf' \
  -F 'strategy=semantic' \
  -F 'chunk_size=1000' \
  -F 'overlap=100'

{
  "job_id": "b2094b38-e432-44b6-a5d0-67bed07d5de1",
  "status": "queued",
  "message": "Document chunking started",
  "quota_remaining": 48954
}

​

Retrieving Results

After the job is created, use the job status and results endpoints to monitor progress and retrieve the parsed chunks:

import requests
import time

def get_chunking_results(job_id, api_key):
    """Monitor job and retrieve results when complete"""
    
    headers = {"X-API-Key": api_key}
    status_url = f"https://visionapi.unsiloed.ai/jobs/{job_id}"
    
    # Poll for completion
    while True:
        response = requests.get(status_url, headers=headers)
        
        if response.status_code == 200:
            status_data = response.json()
            print(f"Job Status: {status_data['status']}")
            
            if status_data['status'] == 'COMPLETED':
                # Get results
                results_url = f"https://visionapi.unsiloed.ai/jobs/{job_id}/result"
                results_response = requests.get(results_url, headers=headers)
                
                if results_response.status_code == 200:
                    return results_response.json()
                else:
                    raise Exception(f"Failed to get results: {results_response.text}")
                    
            elif status_data['status'] == 'FAILED':
                raise Exception(f"Job failed: {status_data.get('error', 'Unknown error')}")
                
        time.sleep(5)  # Check every 5 seconds

# Usage
job_id = "your-job-id"
results = get_chunking_results(job_id, "your-api-key")

​

Expected Results Structure

When the job completes, the results contain the parsed chunks and metadata:

{
  "job_id": "f7e8d9c2-4a5b-6c7d-8e9f-0a1b2c3d4e5f",
  "status": "completed",
  "results": {
    "file_type": "pdf",
    "strategy": "semantic",
    "total_chunks": 15,
    "avg_chunk_size": 847.3,
    "chunks": [
      {
        "chunk_id": 1,
        "text": "Introduction\n\nThis document provides an overview of our quarterly financial performance...",
        "page_number": 1,
        "bbox": {
          "x1": 72,
          "y1": 100,
          "x2": 540,
          "y2": 300
        },
        "confidence": 0.95,
        "element_types": ["Title", "Text"],
        "word_count": 156,
        "char_count": 892
      },
      {
        "chunk_id": 2,
        "text": "Financial Highlights\n\n• Revenue increased by 15% year-over-year\n• Net income grew by 22%...",
        "page_number": 1,
        "bbox": {
          "x1": 72,
          "y1": 320,
          "x2": 540,
          "y2": 480
        },
        "confidence": 0.92,
        "element_types": ["Title", "List"],
        "word_count": 89,
        "char_count": 534
      }
    ],
    "image_dimensions": [
      {
        "page": 1,
        "width": 612,
        "height": 792
      }
    ]
  }
}

​

Advanced Chunking Options

​

zChunk Semantic Chunking

For advanced semantic boundary detection:

data = {
    "strategy": "zchunk",
    "chunk_size": 2000,
    "overlap": 200
}

zChunk uses LLaMA logprobs for intelligent boundary detection, providing high-quality semantic chunks.

​

Error Handling

​

Common Error Scenarios

1. Unsupported File Type: Only PDF, DOCX, and PPTX files are supported
2. File Size Limits: Large files may exceed processing limits
3. Quota Exceeded: API usage limits reached
4. Invalid Strategy: Unsupported chunking strategy specified
5. Processing Timeout: Document too complex or large to process

​

Best Practices

• File Quality: Use high-quality, text-based documents for better results
• Strategy Selection: Choose the appropriate strategy based on your use case:
  • Semantic: Best for maintaining context and meaning
  • Fixed: Good for consistent chunk sizes
  • Hybrid: Most comprehensive for complex documents
• Monitor Jobs: Always check job status before assuming completion
• Handle Failures: Implement retry logic for failed jobs
• Quota Management: Monitor your API quota usage

​

Rate Limits

• Concurrent Jobs: Limited number of active processing jobs per API key
• File Size: Individual files may have size restrictions
• Processing Time: Complex documents may take several minutes to process
• Daily Quota: API usage limits based on your subscription plan

Check your API plan for specific limits and quotas.

​

Use Cases

​

RAG Applications

Use semantic or hybrid chunking to create meaningful chunks for retrieval-augmented generation systems.

​

Document Analysis

Parse documents into structured sections for content analysis and information extraction.

​

Content Organization

Automatically organize long documents into logical sections based on content structure.

​

Search Indexing

Create searchable chunks optimized for document search and retrieval systems.