document-processing

Document Processing

Process unstructured documents into search-ready JSONL chunks using Docling (open-source, runs locally). No AWS credentials or cloud services needed.

Prerequisites

• uv installed (for running Python scripts)

When to Use

• User has unstructured documents (PDF, DOCX, PPTX, XLSX)
• User wants to prepare documents for OpenSearch indexing
• User wants to inspect or evaluate chunk quality

Output

JSONL files at .opensearch/chunks/<index>/<filename>.jsonl. Each line:

{"text": "...", "headings": ["Section Title"], "source_file": "doc.pdf", "chunk_id": 0, "page_number": 1}

The JSONL output can be ingested into any OpenSearch target:

• Local cluster — bulk-index directly
• AOS domain / AOSS collection — via managed-ingestion-service (OSIS pipeline)

Reference

See document_processing_guide.md for the full workflow: processing profiles, quality evaluation, and chunking adjustments.