How to set up Mistral.ai OCR to extract data from a PDF in Temptonic? Teltonik integrates artificial intelligence features accessible to users from simple configurations in automations. To easily delegate writing, summarizing or information extraction tasks with Mistral.ai, optimizing your business processes. Let's look at how the optical character recognition technology of the abbreviation OCR with Mistral AI will allow us to extract data from a PDF format. Also find the tutorial on the same subject which allows you to extract from the image format in our videos. Let's concretely carry out the extraction of banking information from the PDF format together. To carry out this extraction, we will proceed in two stages. And to clarify this explanation, I have split into two types of views the elements that we will find in each of these two automations. For this first automation, as a prerequisite, we will find three columns for the realization of this automation. An attachment column in which we will insert our PDF. A column that will allow us to trigger the extraction request of the PDF information. And finally, a text column that will receive the text format in JSON extracted by the AI. Let's go concretely. Let's create a new automation. Let's call this automation Extraction of PDF in JSON format. The trigger when the tick box will be modified with a double condition. My tick box will be well ticked and my PDF will not be empty. Then In this first part, I select Process a document with the Mistral OCR which will allow me to extract the text. Start by inserting your Mistral API key, choose the type of document that will be processed, in our case it's the columns piece-jointe-rdpj, and where you want the JSON text to be copied in output, so in the column ribtext. I validate my setup and my automation. I close the test window. And let's conduct the first test. I will insert a PDF, my document is present, I'm going to trigger the automation for the scribing of the document. The automation has started. I have here the text of Eason which contains the text elements of my document. I close this window. Let's go to this other view. For this step, as a prerequisite, we only show the columns that concern us, the JSON text column processed upstream by the first automation, a button allowing to trigger this second automation that we will build, which will extract the values from this JSON and send them directly to the relevant fields. Let's go in automations, automation with AI, let's create this new automation and name it. My trigger will be the request to extract the JSON text with a double condition my text is not empty and my triggering button is well checked. Then I select the action images Request to Mistralight Insert your Mistral account API key, your organization ID which is optional. Choose the model Find the information in more detail in the Mistral AI documentation, and insert the role to describe to the AI. And I tell it, from the resulting information of an upscaling of a PDF format into JSON format, you must make extractions. In the question, I tell it you must find the banking information in an oscillation that you must extract in JSON format in the following fields. And I give it the exact name of the fields corresponding to the fields written in the columns of my table. And in the end, I tell it You will find the information in the following field. and I insert the dynamic value of my Dyson text field it should process. I choose the JSON format and I configure each of the values of the JSON in relation to the field where it must be destined. And I continue. like this to address all my fields. I check that the spelling is identical to the destination fields. Once all my values are addressed in my destination fields, I will validate my configuration window. I save my automation. And let's conduct the test, so I select the tick box to trigger the automation and see the extraction of the information in each of the fields. As you will understand, Timetonic makes Mistral.ai available for PDF data extraction with OCR technology. The example is transposable to many use cases. To deduce it, test it quickly!
