PDF Parser with OpenAI
Hi, community! I'm here seeking advice and different points of view about my MVP: GPTParser.
I always struggled a lot to get my invoices organized and had to do plenty of manual work. That's why I came up with the idea of building something that can automate this for me in the most reliable manner.
The idea of GPTParser is to process PDF files into a structured format like JSON, you can ask what fields should be parsed and that's it, no manually recognizing zones or helping the OCR to parse the fields.
The current version is just an MVP but it has several areas for improvement:
- Output type validation
- Have an API so people can use it from the code.
- Add support for more input document formats like docx, or doc or even HTML
- Add some way of controlling the number of requests and don't ask the user for his OpenAI API key
etc ..
Do you think this could be a SaaS or a monetizable product?
How would you market this?
Any feedback you can give is welcome!
You can try the app here: https://gptparser.com
Love that your trying to improve this space. Quite a saturated market though, for structured data (aka bank statements, invoicing) into the accounting system e.g. MyobCapture is FREE for their user base.
The email inbox is filtered for inbound PDFs triaged parser and creates transaction automatically with PDF (source document) attached. Auto-filling comments in the trasnaction in the accounting system is where it needs improvement. Multi-page docs are the trickiest. After helping SMEs digitalize for yonks it is always about the making it a complete "hands-off "process. MYOBcapture is just one, there are even better ones back 3 years ago went I did a deep dive. The gap I am seeing is ability to capture unstructured to eliminate tedious-nesss e.g. bucket of parts in my garage some with IDs on part, some with packaging info, and I want to make a parts list from that super easy and smart (aka enjoyable to use).
If you desire more ideas/insights on this, please lmk as done heaps of comparison + deep dives in this space.
Thank you for your detailed response! I didn't know MyobCapture and it seems really similar to what I'm doing. I will dig deeper into your suggestion about auto-filling comments, thank you again!
A pleasure. I remembered the other pain point -- TOC. Textbook style TOC are complicated to parse. 3 column layouts, 2 column layouts. People need it to create the bookmarked TOC quickly in the PDF. They extract the PDF pages, parse it, proof read the text (which I now use chatgpt for), then automate the bookmark creation using a python script. I do this at least 10 times a week and its still painful. Fix that for a mac user, and I will pay gladly. There are many average free solutions in this space. Tabular is my preferred(https://github.com/tabulapdf/tabula
This post, is one of the best researched piece of structured data in PDF to text. They have not automated the process but are totally up to speed https://github.com/aminya/tocPDF
Alterative: If you are into parsing, there is a much greater need gap. Image with unstructured text (live text or SDK) to record based system (e.g. parts list, issue list). Not found any solid solution but hopeful soon will as new iphone live text capability and mobile shopping shift. Sparkscan SDK released mid jan 2023. Many other use-cases as it enables the crowd source data capture shift.. 'Why type if you can snap' LIke SnapSendSolve app but freely available for citizens. Citizen Scientist for the ocean cleanup programme is the single app . no mainstream app yet. https://theoceancleanup.com/research/citizen-science/citizen-science-map/