REVOLUTIONARY!!!!! UiPath Document Understanding and Generative AI - Invoice Data Extraction

so in the previous video we have seen how we can classify documents using a single activity in uipath where we only need to configure the labels and their description in this one we're going to do something more impressive which is data extraction using the same logic we're going to have labels of the fields that we want to extract and their definition and then we're going to send that to action center to be validated by a human agent the example of the documents that we are going to use is invoice so we're going to use the same labels to extract information the same information from multiple invoices that are different so if for example we are working with multiple client invoices we can sell extract the same information using one single activity so you want to stick around for this one and if you are enjoying this type of content drop a like And subscribe it's much appreciated and let's jump to my screen okay so we're gonna start from where we left off last sign so we have this process that I've already created if you haven't seen the first video just look my channel up and then watch this video don't miss out your iPad documents understanding and generative Ai and you're going to know exactly how I created this this process so in today's video as we said we're only going to see invoices and how we're going to extract data from them so in my folder where I have all the documents I have all of these invoices that we are going to be extracting the same information from them so the number of the invoice the address the total amount etc etc so these are the things that we want to extract from these invoices for example here this is the number of invoice this is the total amount this is the due dates or generally the date and this is the address for example that we want to extract let's go back to uipath and here we're gonna see what we have done so we basically go through all the documents in the folder using this activity then we get the I file using the activity path exists from here then we're going to classify the documents as we have said then in the switch we copy the file to the correspondence folder we're not going to do that today we only want to extract the data we don't need to copy anything into any folder so yeah so let's go back to invoice the classification of invoice and let's delete the copy file and now what we are going to do is we're going to use the activity extract documents data so we're going to use this activity extract document data and we're going to put it in here in the project we are going to choose predefined then we are going to choose generative preview in here and extractor and in the inputs file we're going to put the I file and now we have to make sure what are the information that we are going to be extracting and we have to pinpoint an exact information so for example here I will have sometimes a build tool sometimes I will have a ship too so I have to actually go and know exactly what I'm going to be extracting okay so we're going to see the number of invoice that's the first thing the second thing is the total amount we want to extract that as well then do the bill to address so we want to extract the bill to address and the last thing that we want to extract is the date so the date of the invoice okay so we have the four fields that we want to extract and now let's go to a UI pad and start working basically on the fields so as we have seen in the classification we use a key and a value it's going to be the same thing in the project here is going to be the classification of the documents itself and but here in the extraction it's going to be the field of the extraction so here for example the first thing we said is going to be the invoice number so let's have an invoice number in here let's add three other entries then let's have the total amount or just total then after that we're going to have the Bill to address and then the dates of the invoice and in the value I don't want to do that myself I'm just going to go to chat GPT and get the definition so let me go to chatgpt and and dates okay so here the invoice number this is the definition let's put it inside here total amount this is the definition let's copy let's put it inside here okay the bill to address and the date so I have missed d okay so let's save this so we're gonna go to properties and then here we are going to click on Ctrl K to create a new variable and let's call it extraction invoice a good thing about Ctrl K when you use it is that you don't have to deal with the variable type so we don't have to deal with the variable type because sometimes you have to look for the valuable type in browse for type and it could be a hassle if you don't know the exact name or sometimes two variable types can have the same name so always use Ctrl K to control the to create the variable inside of your activities that's good so now we have extracted the data we want to make sure that our data is correct and what we do is the message box so let's have a message box just to visualize the data make sure that it's correct and then after that we're gonna go and add the data to action center to be able to validate it visually so here let me go to chat GPT again and vb.net I don't want to be writing the syntax so chat GPT can write it for us and it's good so here I have this I will just copy it and then I will open it here and I will I will uh I will paste it and inside of here what I will do is I will have the extraction invoice okay it doesn't want to show me invoice dots data dots and here I will have Fields so for example let's have the invoice number and let's highlight that to string and then let's copy this and let's change the fields once you once you create the extraction invoice variable the invoice number the dates and all of the other fields are going to be inside of the data they are going to be created inside of the data so you can extract them easily so here I'm going to copy it here then inside of the third string and then inside of the fourth string and I'm gonna go back here the second one and I'm going to choose the invoice number 2D to the total then the third one I am going to change the invoice number to the Bell to address and then the last one is going to be dots dates and now I will be able to visualize them okay so now we can run the process so let's click here and click on run file we are positioned on the invoice so we can see if it will be able to basically extract the invoice as you can see it's not the best quality but still I hope the OCR is gonna work correctly so as you can see here okay the uh the number is correct the total amount is correct uh the address is correct so we have the full address the name of the company and then the address in here that's really good let's read actually they're really impressive because we're having another address in here but it's recognized the address that we want so this is actually very good from the first try sometimes you have to have to work with your prompts in order for you to get it right especially if you have two similar information and this one is not even labeled so that's really good and the date is June 19 2019 is correct as well we're gonna see the second one and then we're gonna sub the process so this position on the second invoice this is a bit more complicated as an invoice actually it's not going to be this one because this is document three this is the second one so it's an AWS uh invoice and as you can see here I have the full invoice number so it's recognized even the there is a huge gap between them it has been able to recognize it and the total amount is 411 that's really good uh the bill to address it's really good as well it's Hong Kong all of it that's really good and August 13 2014 which is the address where we can refine that August 3 I I can okay so our amounts due on August 3 2014. okay that's really good okay so that's that's everything I I can see that everything is correct so we can stop the process here and go to the second parts of our process which is basically sending everything to action center so let's delete the message box and let's have another activity called let's have another activity called create validation task that's good and here we're going to have the extracted data so extracts in invoice and let's just give the uh a title for the action and it should be validation of invoice and here we're gonna have plus and current file name that's me so we can know what we are validating which invoice we are validating so let's run the process again of course I suppose that you are connected your robot is connected to a cloud instance so if you go to assistance your iPad assistant it should be connected in here and you should basically have all the rights necessary so this is not a video about all the rights in order for you to be connected to a cloud instance and all of that and having the action center set up this is a video only about the validation so I suppose that you already have that if you want me to make a video dedicated for that just tell me in the comments I can make it no problem okay so the process has finished that's really good 44 seconds for five documents now let's go to cloud.uiped.com and in here I will go to action center so where is it yeah it's here and this is apps integration actions so we're gonna go to action center and inside of here if I go to inbox and I go to unassign I will find that the five documents has been sent to the uh to the action sensor and now I can validate them and the best thing about this is that I can see an instance of the document inside of here so I can actually validate it just by clicking on the elements but that's really cool and this is the best thing about the integration between generative Ai and documents understanding is this user friendliness you cannot have this user friendliness elsewhere even though that you can use I don't know uh even if you're using other tools you're not going to have this type of sophistication in the validation process okay so here we have the first invoice so validation of invoice documents one and as you can see here everything seems to be correct and we can see the degree of confidence now one thing I would say about generative AI it's always 100 confident it's like the most confident thing that you have ever seen and this is just an image of what for example chat GPT is if you asked chat GPC about anything even if it's giving you the wrong answer it will still be so sure about itself so this is the difference between generative Ai and other and machine learning models at least they have different confidence level in here it's always a hundred percent so that's that's I think that's the only drawback of generative AI is the fact that it's always so confident and it shouldn't be let's see the second documents if everything is correct we've seen that so this goes to the third one so okay so here I don't like that as you can see here uh it took invoice as well and I don't like that I don't like the fact that it took invoice as a invoice number as well I just want the number I don't want invoice inside of the invoice number field so to be able to change that I just need to assign it to myself so I will click here and I will click to assign to self and now I will go to pending instead of unassigned I will go to a pendant okay so now I will go to the uh the task that I've just assigned to myself and here I will click on the value here and I can basically change it so I can delete this value and only keep the number and here I will uh yeah so that's good so you just need to do it and it saves by itself we will click submit at the end here it's good the value that we have here it's good as well since it's the only address in here even though that we don't have a bill to address and the date is good as well the date is good great so here we're going to click on submits continue and Save and as you can see here it will go to the completed this is an old one yeah so it will go to the completed and as you can see here it's it's really good so so everything looks correct and this is how you can basically validate the data if the model hasn't been able to correctly predict it so in this case it wasn't perfect but still it has been able to detect the right uh invoice number it's just that it has another thing with it that's that's not ideal so it was not 100 failure so so that's basically it you can assign everything to yourself and see if it's good and then validate it and if you have enough confidence in the model and it has been able to predict a lot of a lot of invoices correctly you don't have to assign it to yourself you can basically use that data and write it into a database or write it in an Excel sheet or whatever output you want to have

Channel: Reda Marzouk

Views: 4,513

Keywords: rpa, uipath, documentunderstanding, invoiceprocessing, dataextraction, ai documentautomation, roboticprocessautomation, intelligentdocumentprocessing, automatetasks, businessprocessautomation, automationrevolution, learnrpa, uipathtutorial documentunderstandingdemo, invoiceprocessingdemo, robotics, machinelearning, artificialintelligence, deeplearning, computervision, naturallanguageprocessing business, finance, accounting, procurement, supplychain, humanresources, marketing, sales education

Id: kg1phmgC2Ko

Length: 16min 22sec (982 seconds)

Published: Tue Aug 22 2023