3 Ways to Extract Data from PDFs with Microsoft - ChatGPT, AI Builder, Syntex

[@MCSurvivalMode]

If you want to extract specific fields using ChatGPT, use compose but make sure you have a delimiter like a semicolon and grab the text in between using Split. split(split(outputs('Compose'), 'VendorName: ')[1], ';')[0]. Please let me know if anyone has a better way.

[@andrewhess123]

Nice love the idea, but for some reason it kept making random special characters for me. I feel like if we just wait for 1 more release of a new LLM like GPT5, we can get it perfect! Other options could be like ask it to respond in the format of JSON? But love the idea of asking it to use the delimiter.

[@saeedsm57]

100% Useful! Thanks Andrew!

[@andrewhess123]

Thank you Saeed!

[@kyle.garrett]

Thank you. Very helpful.

[@andrewhess123]

Thanks for watching Kyle.

[@beecee9207]

Thank you Andrew this is amazing stuff!

[@andrewhess123]

Thanks for your continuous support bee!

[@waleedkhan2500]

Awesome tutorial again Andrew! Have you considered working with the assistant API and bringing forth vector stores for training or GL tracking? Would love to see how to make that work with Power automate

[@andrewhess123]

Hi Waleed! That would be amazing stuff to do! Got some learning to do!

[@Sophie-fh3lh]

Hi Andrew Hess, Do you have to pay a subscription to ChatGPT or AI Builder?

[@andrewhess123]

Hi Sophie! ChatGPT API is based on consumption. I put 10$ in and dev'd for months and only spent like $2.00. I believe ChatGPT has a new model 4o-mini and the price dropped significantly. AI builder comes with credits with premium Power Apps, once again it's based on credits. So if you have a few users with Power Apps premium you can have a bunch of credits already. If you don't have Power Apps premium at all it does cost extra. There is another way not mentioned I've been testing in Azure Document Intelligence and you can currently do 500 pages for free per month.

[@Sophie-fh3lh]

@@andrewhess123 Thank you Andrew!

[@kyle.garrett]

What if you just wanted to dump a mix of different documents into a single location (e.g. a OneDrive folder), and have it automatically separate the documents into their respective types. Can these tools do that? Perhaps ChatGPT would be best at “sorting”?

[@andrewhess123]

So you have 2 main options I can think of, #1 is SharePoint premium. It's going to be able to do this very well. The other option is ChatGPT with the api you just have the AI model look at the document and attach metadata to a column and have your sorting built into the columns. I think SharePoint premium is the better options it's worth looking into. techcommunity.microsoft.com/t5/sharepoint-premium-blog/introducing-sharepoint-premium-the-future-of-ai-powered-content/ba-p/3981076

[@starbusttweety]

In power automate cloud the model is taking too long to train. The internet is fine. What should I do?

[@andrewhess123]

AI Hub with AI Builder? Depending on how many fields and how many documents its trained on, it could probably take several hours.

[@LeAn-rl5yn]

Hi @Andrew, I really enjoyed your video! I have a question. Since Microsoft 365 Copilot offers similar functionality, such as reading attachment files and performing tasks based on a prompt, why don’t you use 365 Copilot? Does it not support APIs? I’m new to Azure and Power Automate, and I want to send a PDF file from an external source to Azure via REST, have AI read the file, generate data according to a prompt, and then send it back. Could you give me some advice on how to approach this?(Fully automated, no manual work even file upload or create promt) Thanks a lot!

[@andrewhess123]

Hi Le An, sadly Copilot 365 is still not available currently today in the GCC (Government Consumer Cloud). Also 365 Copilot is using Chat GPT as it's back end. The main goal is to completely automate it like you said, fully automated. Azure with REST, is probably the best way to do it. Using Azure Document Intelligence which used to be called Form Recognizer with a neural model was easy to set up with Power Automate. You would go to Azure, set up and train a model, then using Power Automate to call on that model and send you back data. I do think I should make a video on this. Would be great!

[@andrewhess123]

Hi Le An, I created a new video today you may want to check out on extracting data with Azure Document Intelligence: kzbin.info/www/bejne/kIHImXuZoa2DhZY

[@LeAn-rl5yn]

@@andrewhess123 niceeee, thank you!

[@YOUSmegmaster]

The flow returns token limit exceeded for way #1 and all my attempts to limit token usage are failing. Any steps that may not be in your video or ideas why this would be happening. Thanks in advance.

[@dns2557]

Hi Andrew, how to extract pdf to excel file using AI builder? I mean how to create a new file instead add a row into my template? I have been adding actions wrongly i believe

[@DavidPhilip-yk2yl]

Also, the link to the video on creating a custom connector to ChatGPT-4o isn't in the description of the video. I don't know if it is possible to also create a custom connector to google Gemini AI using power automate. Thanks a lot. You are the man!

[@andrewhess123]

Hi David, Thanks added it in the description. Yeah I'm sure it should be possible to use Google Gemini AI, and if it's not available yet it will be in the future.

[@hammadyounas2688]

Great Work Sir, Did it work for get attachment content?

[@andrewhess123]

Hi Hammad! Should work just the same from get attachment content. ChatGPT needs a base64 format where AI Builder did not need it. You could also push an attachment to SharePoint and have SharePoint Syntex run automatically on all documents.

[@hammadyounas2688]

@@andrewhess123 Thanks Sir. I will test it. Hopefully it works

[@starbusttweety]

In power automate cloud the model is taking too long to train. The internet is fine. What should I do?

[@DavidPhilip-yk2yl]

Awesome stuff per usual. Can you please make a video using Azure document intelligence for custom PDFs and Power automate to extract the data and an automation that sends the data to a SharePoint list. I tried doing this using the free plan for 500pages per month, but I kept hitting road blocks after i had trained the model and tried to work with the results in json format

[@andrewhess123]

Yep, I actually recently did this. Totally worth a video! Writing down now.

[@andrewhess123]

Hi David! Your video is ready 😄 kzbin.info/www/bejne/kIHImXuZoa2DhZY