Cardinal

Hey everyone – Devi and Jianna here 👋🏼

We met while studying at MIT and Harvard, and our last company processed utility data at scale. The biggest bottleneck? OCR. It was terrible - especially for complex documents where preserving structure was critical. After years of wrestling with these OCR limitations, we decided to build something better. That became Cardinal.

📃 The Problem

Most enterprise knowledge is trapped in PDFs and other unstructured formats -  medical forms, contracts, invoices, insurance claims, financial statements, packing slips, etc.

OCR is not a solved problem:

• LLMs hallucinate and can fabricate data
• Annotations, checkmarks, and handwritten notes get ignored
• Tables, charts, and images are dropped entirely
• Layouts are mangled, breaking document structure and making the output unusable for code, retrieval, or downstream automation. Preserving the semantic meaning of the document is nearly impossible.

🚀 Our Solution

Cardinal has 2 finetuned models. The first model, Document-to-Markdown, accurately converts even complex documents into clean, structured Markdown. The second model, Markdown-to-Code, then turns that Markdown into ready-to-use HTML, preserving the original formatting and hierarchy exactly.

Put simply, we can help you:

• Reliably extract structured data from the most complex PDFs
• Preserve exact layouts for retrieval, search, and downstream LLM workflows
• Convert charts into data, summarize images, and parse complex tables
• Run natural language extractions for any field without brittle regex or templates

Full Demo video: https://youtu.be/RouYM1cKGXI

🔗 Demo
Try it here - we’d love your feedback!

🙏🏼 Our Ask

If you or your team works in enterprise and deals with complex document workflows, we’d love to chat.

📅 Here’s our Calendly: Book time here or email us at team@trycardinal.ai.