Markdown Output

The goal of document parsing is “clean, structured Markdown” that works well for:

• search and retrieval
• RAG ingestion
• downstream structuring (table extraction, section chunking, citation pipelines)

Excel output

Each worksheet is separated by an H2 heading and rendered as a table.

## Sheet1

| Name | Value |
|---|---|
| Item 1 | 100 |

Word output

Word documents aim to preserve headings, paragraphs, lists, and tables to keep reading order natural.

PDF output

PDF output depends on layout and reading-order inference. Common structures include:

• section headings (h1/h2/h3)
• paragraphs and lists
• tables (when reliably recognized)
• formulas (when representable)

Retrieval-oriented tips

• chunk by section first, then by token budget
• keep source URL and page/section metadata in your pipeline if possible
• treat bad outputs as regression samples and iterate

Related:

• Tables & Formulas
• Errors & Troubleshooting