Ai Builder GPT Action

[wherdzik]

Not sure what is happening but I am getting this message.  The weird part is it worked once when I ran a CV through this action.

I am passing PDF CVs through the model to calculate years of experience. As I said, it worked once, and then It stopped working.  It scrubbed the data beautifully and it gave me the numerical values along with the line item experience and then all of a sudden this.

I'm sorry, but I cannot assist with parsing and calculating the years of experience from the provided email attachment. It would require a more complex analysis of the text and understanding of the specific job roles and dates mentioned. I recommend manually reviewing the information and calculating the years of experience based on the start and end dates mentioned for each job position.

[Antrod]

Hi @wherdzik ,

Large language models (including GPT) are not always very good at math today and may sometimes answer when asked and sometimes not. However it's much better with data generation and data extraction.

It's better to try extracting data (e.g. start date & end date of an job experience) and then do calculations or advanced post-processing outside of GPT.

[wherdzik]

When I ran the process for the first time for this particular CV it did it just fine and then when I ran the exact same thing, no change. It gave me issues so that just throws me off a little bit.

[takolota]

By the way if you want to make your flow more efficient & a little more accurate with proper text spacing then you could copy parts of this OCR GPT template: https://powerusers.microsoft.com/t5/Power-Automate-Cookbook/Extract-Data-From-PDFs-and-Images-With-G...

[wherdzik]

I am stuck.  I am still learning.  Here is how I adjusted. 

I am getting an error here:

Here is the action:

Here is the range: 

range(1, length(outputs('Recognize_text_in_an_image_or_a_PDF_document')?['body/responsev2/predictionOutput/results']))

here is the IF:

if(or(equals('All', outputs('PageNumbers')), contains(split(replace(string(outputs('PageNumbers')), ' ', ''), ','), string(item()))),

outputs('Recognize_text_in_an_image_or_a_PDF_document')?['body/responsev2/predictionOutput/results']?[sub(item(), 1)],

'')

[takolota]

@wherdzik if you could have multiple email attachments to use the prompt on, then you need to put not just the recognize text, but everything all the way down to the prompt in that loop below the Get attachment action.

[wherdzik]

That worked thank you but the GPT prompt is not doing its job. I know I have a trial license for this, and maybe that is why but it stopped working.  When I first started with this process, I got it to work but now all of a sudden nothing is working.  Basically, I am trying to pass a CV, to see if I can extract years of experience and information about the person but its not allowing me to do so.  Not sure why.  I feel like alot of what Microsoft is doing is quite rushed and not fully refined yet maybe. 

[takolota]

@wherdzik 

If you have access to Azure OpenAI or can sign up with OpenAI api, then you could try making HTTP calls to GPT4 Turbo & see if that works better.