What breaks in document AI

Batches, timeouts, slow ingestion

19 of 19 quotes

• “Two public PDFs made the ingestor die 49 times in about 3 hours (33 OOM kills, 16 panics)”

• “We had to move off Textract because our documents cannot leave our own cloud.”

• “causes a "Timed out waiting for result of task XX" timeout error when mineru processes larger PDF documents, leading to task failure.”

• “I’m working with large amount of papers so it’s not possible to audit each separately.”

• “After batch uploading documents to a newly created dataset, the metadata of all documents across all existing datasets in the account was unexpectedly and completely cleared.”

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

• “From my experience, Docling without VLM does a good job, but it's too slow to process large volumes.”

• “silent failure caused by router timeout when processing multiple pdf”

• “We can not process large PDF files in pure CPU mode, because the program exceeds the maximum allowed time, causing the normally running progress to be forcibly terminated.”

• “When using docling to convert multiple docx files in a multithreaded environment, failure files may occur.”

• “An extraction job was submitted successfully but never progressed.”

• “PaddleOCR再运行10分钟左右，ragflow报错read time out，但PaddleOCR还在正常运行。”

• “the inconsistent timing in tier="agentic" makes SLA guarantees and timeout tuning impossible”

• “docling is taking more than 20 mins to process the document. This feature is required in my application on frontend, so user experince perspective 20 mins are long time”

• “I had a bit of a nightmare integrating a docling pipeline earlier this year. Docs said it was VLM-ready, which I spent lots of hours finding out was not true”

• “when dealing with large PDFs (over 500 pages), the document conversion process takes too long.”

• “I make the api request to docling serve and poll for response and it always times out (currently the max timeout is set to 2400 seconds).”

• “After the batch import fails, I've tried parsing each document individually, and it parses successfully each time.”

• “the entire process of converting to Markdown takes up to 10 minutes.”

• “V0.20.4 DeepDoc batch upload parsing speed is slow (2000 files), consistently below 1%”

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