Monkt

Hey everyone, I appreciate your support; the feedback I've received so far has been amazing. You made my day on January 1st! :D I’ve added some extra capacity since we experienced some spikes in traffic over the last hour. Everything is running smoothly now. I can offer a lifetime discount for anyone who has experienced technical difficulties! Just send me a screenshot through my contact channels. Thank you once again!

@julia_zakharova2 Hi Julia, Thanks a lot for the nice words, appreciated!

@sentry_co I guess the author want to express the transformed files can be understand by AI System(such as ChatGPT) directly

@sentry_co Hi André, Thank you for the question. By "AI-ready" I mean that the files are in a format suitable for prompting large language models or for creating knowledge bases that can be utilized by various AI tools. Additionally, I want to "send a message" that this tool is primarily designed for AI practitioners, although it can also be used for other purposes.

@kkofficial Hello Kunal, Yes, but in a few months. I appreciate your interest.

@renchu_song Hey Richard, Thanks for the nice words; I appreciate them! Regarding your question, the conversion to Markdown is happening purely on our server. For JSON schemas, we are using LLM with a big context. We are not keeping the original files and training on that. The storage of the derivate file is encrypted. Hope I answered.

@tibelf Hi Leo, Can you share with me some example problematique cases? I would take a look. You can reach me via X, LinkedIn, or support channels. In short: transfer to MD should work without any troubles. Transforming to JSON could be further improved in my opinion, but for relatively short docs -> should work great. Pre-defined prompts could help doing some more complex operations, like summarization, translation, etc. Thanks.

@krazygaurav93 Hey Gaurav, Thank you for the nice words. I have published this article with more information: https://medium.com/@simeon.emanu... Will record a few videos in the next two days. I think the API can bring a lot of value, so I will first make a video overview of this part.

@simeon @simeon_emanuilov I see you showed "MarkItDown". It's good but as you correctly pointed out it's not scalable. Also using LLM to parse document is just non sense, it becomes expensive, slow and non scalable. However you have not disclosed your solution there, I understand why. In case you want to brainstorm on the solution for further improvement let me know. Well this is a very common scientific problem you are trying to solve. I personally like "OpenParse" but it has limitations. I then started to use "Marker" which is really good but is commercial and is open for research only. I think OCR is the right way to go, I am thinking of training my own model on huge PDF formats to solve problems like headings, two-side pdf, multi-page tables etc. Best of luck.

@jalcantara Thanks Jorge. The main difference is our focus on ML/AI document processing rather than general web crawling. We preserve complete document structure and semantic relationships for Markdown exports, support custom JSON schemas, and optimize output for ML pipelines.