Hi @lelloba
You are my helper regex Guy 
Could you help me to extract the values from the PDFs (files) from this folder?
The values will change.
I need the pattern.
All PDFs have the same layout.
I can’t get the right values.
Reading_GNRE_Workflow.knwf (3.1 MB)
@Felipereis50 you can take a look at this article and the example to define areas of a pdf to extract. To get the definitions you could employ a LLM.
Here is one example how to extract information from a bank statement which were less suffered with a local LLM and GPT4All:
A basic python version might be faster.
Hi @mlauber71
You know you’re at a hard level, right? 
But I’ll take on the challenge…
I’m following the steps of the workflow: “GPT4All - Extract Information from PDF Bank Statements into JSON – mlauber71”
Well, help me if you can:
Step 1) I accessed the website GPT4All
GPT4All
Step 2) I installed the application
Step 3) I’m stuck at the step for the NODE: “prepare GPT4all model path”
Where is the .bin path?
In my PC, I installed the app in this path
C:\Users\felipesr\gpt4all\bin
But, I don’t know if I need to download an external file in this location from the app and then adjust the path.
Plus: Do I need to pay anything to use ChatGPT4All in this workflow?
@Felipereis50 you can read about the setup in this article
Hi @Felipereis50 .
I suppose you need to download the model first.
Thank you. It worked
@hmfa
@mlauber71
I ran some tests and couldn’t get it to work, has I expected.
I even tried using ChatGPT but didn’t get a good result.
Here’s what I’m thinking:
When I wrote the prompt, I did it by looking at the PDF.
Let’s analyze the screenshot of the PDF.
When I wrote the prompt, I said something like this:
Extract the values from the PDF by considering the header I’ll write in quotation marks and the value right below it in square brackets.
Example: "UF Favorecida" [AL]
(I passed all the fields I needed.)
However, looking at Knime, and I imagine ChatGPT works the same way, the PDF data is converted into text—and here’s where the value order gets mixed up.
Take a look at the screenshot of Knime’s output.
This way, I can’t get a satisfactory result.
| Column1 | Column2 |
|---|---|
| valuespdf1 | valuespdf1 |
| valuespdf2 | valuespdf2 |
| … | … |
I’ve seen a feature in Power Automate (Microsoft) where you map data based on the PDF layout. You click on the image and save the values into a variable.
@Felipereis50 I tried a few things. The LLMs were not very satisfactory. I think if you cannot come up with a regex you best chance might be to find the right positions on the sheets and try one of the Phyton packages. I tried that also with LLM but the results were not very encouraging.
In think Microsoft has some professional (paid) services to extract such data.
Thank you for the feedback. @mlauber71
Regex is definitely the best solution.
I would need to identify the “words” that come before each value I need.
For example: I need the value AL, which comes right after the word GNRE. So, I would need to identify GNRE and fetch the next two characters, and so on.
Anyway, I’m not an expert in Regex. I find it quite complex.
I bought a course on Udemy today Let’s see if I can learn something.
And about Microsoft:
I don’t remember very well, but I think I could use POWER AUTOMATE DESKTOP.
But I also need to learn it to do a LOOP through each file.
There’s so much to learn. It drives you crazy 
Thank you very much.
The important thing is that I learned how to configure GPT4all.
I’ll mark this topic as solved.
If there is no restriction on using any other software other than KNIME, then I’d suggest using Tabula for this
I’ve used it recently to extract data from the National Achievement Survey reports (https://nas.gov.in/download-data-district-wise-2017). Each of the reports have the same structure so using tabula would be great for these kind of tasks
Also, could you share the KNIME workflow you’d shared earlier in this thread as a screenshot?
Thanks
Great.
I’m checking the Tabula software.
About the workflow, here it is:
Reading_GNRE_Workflow.knwf (3.1 MB)
I used “Tabula,” and it is a good option; however, it does not perform a loop for each file.
You can upload multiple files, but you need to click one by one to download the data.
It’s not practical.
But the concept is good.
@Felipereis50 you could try and take a look at these examples also using tablua and multiple PDFs. Maybe you can share your approach that worked so it is possible to adapt it.
This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.
