DocCrush
Convert any document into clean Markdown for AI.
14 followers
Convert any document into clean Markdown for AI.
14 followers
DocCrush converts PDFs, Word, PowerPoint, Excel, images and more into clean, structured Markdown. Built for developers, AI applications and RAG pipelines. • PDF → Markdown • Word → Markdown • OCR for scanned PDFs • Tables preserved • Images extracted • Batch processing • AI-ready output
Tried it on a messy scanned PDF and the table extraction came through surprisingly intact, saved me a ton of cleanup before feeding it into my RAG pipeline.
@kemaltzly Thanks for giving it a try! 🙌
Great to hear that the table extraction helped reduce cleanup before your RAG pipeline. Handling messy scanned documents is one of the areas we’re focusing on.
Appreciate the feedback — it’s exactly the kind of workflow we want to make easier.
honestly looks solid for our RAG setup. one thing that would be super useful is a way to preserve the original page numbers or section headers as metadata in the markdown output, since right now our chunking loses that context. maybe a config flag for it?
@rhastekkes51150 Thanks! That’s a great suggestion.
Preserving page numbers and section headers as metadata is definitely something we’re considering, especially for RAG workflows where chunk-level context matters.
A configurable option (e.g. include_metadata) sounds like a good approach so users can choose between clean Markdown and RAG-friendly output.
I’ll add this to the roadmap. Really appreciate the feedback!
Super useful for RAG pipelines. One thing that would make this even better is the ability to preserve original heading hierarchy and add frontmatter metadata (title, author, page count) automatically so the output drops straight into documentation sites without extra cleanup.
@oykuener77260 Thanks! That’s a fantastic suggestion.
Preserving the original heading hierarchy is definitely important, especially for both RAG pipelines and documentation workflows.
Automatically generating frontmatter (e.g. title, author, page count, source) also makes a lot of sense and would allow the Markdown to be dropped directly into tools like Docusaurus, MkDocs, or other static site generators.
I’ll add this to the roadmap. Really appreciate the detailed feedback!
honestly this looks super useful for RAG pipelines, but it would be a huge win if you could add an option to output with YAML frontmatter included automatically, so the title, author and page count get bundled right into the markdown file for easier chunking downstream.
@tunahankb0d Thanks for the feedback! 🙌
Great point — adding YAML frontmatter with document metadata would make the output much more RAG-friendly and eliminate extra cleanup steps downstream.
We’re definitely considering this for future updates. Appreciate you sharing your workflow!
Finally tried this on a messy scanned PDF and the tables actually came through intact, which is more than I can say for most converters I've used. Set up a quick batch job through the CLI too and it just worked.
@mucahit_sa21533 Thanks for trying it out! 🙌
Really glad to hear that the scanned PDF and table extraction worked well for your workflow.
Handling messy documents (especially scanned files and complex tables) is one of the areas we’ve been focusing on, since it’s often where traditional converters struggle.
Great to hear the CLI batch workflow worked smoothly too — we’re continuing to improve the automation side for larger document processing pipelines.
Appreciate you sharing your experience!
Really solid for pulling clean text out of messy docs. One thing that would make it way more useful for my workflow though is preserving the original page numbers or section headers as metadata, so when I chunk things for RAG I still know where each snippet came from.
@salih416617 Thanks for the great suggestion! 🙌
You’re absolutely right — keeping page numbers and section context during chunking is critical for reliable RAG pipelines.
Adding metadata preservation (pages, sections, source info) is something we’re considering to make the output more RAG-friendly.
Appreciate the feedback!