What breaks in document AI

Scans and OCR quality

12 of 12 quotes

• “I've made a personal doc tool based on Mistral OCR API at first, and then switched to Gemini Flash for the same price inline, or half price using batch API, and the difference is night and day. Mistral doesn't even come close on anything non-trivial.”

• “the hallucinations in 4.0 were so egregious (just completely making up new sentences in the middle of a page) that I knew I can't trust Mistral OCR on its own.”

• “Docling is good but it missed, especially for complex vector graphics.”

• “I was ingesting a lot of old maths pdfs which have specialist notation that wasn’t always getting correctly ingested and as they were image pdfs it was coming in as pure garbage.”

• “When using docling to parse scanned PDFs, it was found that some paragraphs turned into meaningless letters”

• Hit this one too? We are building for it. Join the waitlist.

• “So far I have tried paddle OCR and Google's OCR, the results were underwhelming with hallucinations and missing details.”

• “I used their OCR against a few hundred page PDF that was printed text but missing the OCR. It cost me $5 and was useless, it did worse than tesseract.”

• “when a centered watermark is present on every page, the extraction becomes noisy and unreliable.”

• “The JSTOR-style scans have "dirty" text (e.g., 'f' being read as 's', or broken ligatures) that breaks embedding semantic consistency.”

• “Even with modern OCR (I used Mistral's here), and a book with limited formatting, it's funny how hours of touch-ups are required”

• “There are no spaces between words.”

• “I am still facing challenges related to layout variability, inconsistent OCR accuracy, multi-column alignment issues, and difficulty in extracting PHI embedded within headers, footers, tables and annotations.”

• Hit this one too? We are building for it. Join the waitlist.