so welcome to AI Academy my name is l Sanchez I'm an outbound pm on the platform AI side and before we get started I just always like to mention that we have additional resources for you there is an AI and intelligence community on the service now Community Forum you can access with an Easy Link it's S.W works on that Forum you can ask questions we have experts and other customers and partners uh joining into uh answering your questions we also publish articles and it's a becoming quite uh populated with a lot of articles on the topic of the different AI products so I I recommend that you check check that out whenever you want information and the the great thing is that it's available all the time so you don't have to to wait to get an answer in addition to that uh we have the those recordings that are posted to the YouTube channel of the now community and you can find a previous recording of that those academies as well as the other ones and all the content that's created by the now Community different teams so very helpful forum and if you want to get the latest news you have a YouTube account you can subscribe to the channel so you get updated Whenever there is new content um and for today specifically I want to mention uh the document intelligence AQ Forum that was just published recently so I'm going to put the link in the chat uh it's easily accessible when you go on the community but just want to mention that because it's pretty it's uh fairly new we gathered a lot of the questions that we got on the field uh from from customers Partners uh around you know the frequently asked questions on on docking and we just compile that into document that's easy to consume so you get all the answers for your questions without having to wait for somebody to to answer you so I again recommend checking that out and other resource that uh we want to mention today uh the the AI team just launched a new course on now learning the course is is called get started with AI Solutions it's aimed at anybody in your company that has to do with AI and service now so AI process owners AI admins even just if you want to be an AI champion in your company and it brings you really all the the basic uh understanding of what's AI how to deploy it and now to make AI driven decisions across your company so it's uh not uh technical course on how to implement products but it's a foundational course that's very helpful to to help you understand and access which AI Solutions you need to to be able to deploy them so you want check that out on that's on now learning as always um we're going to start just spending a few minute looking at the safe Haro notice in those sessions we tend to only talk about what's available in the product how it works uh but sometimes if you have a question and we want to answer you and give you a sneak peit of what's coming in the future we might mention something that's not available yet in that case please exhaust judgment and don't make a purchasing decision based on that information and always check back with your account team on what's available in the product at the time you purchase it if you are if you are new to AI Academy here are some Basics this is and sessions that we bring for you so we bring you fresh ideas to give you a better understanding and practical guidance of our AI products those sessions are recorded and then we post the link on the YouTube channel that I mentioned previously uh but if you are here with us live today U you can enjoy that content and you can also ask questions so use the Q&A panel in that Zoom webinar to ask questions and we are here to answer those today and so that would bring us to the topic of today which is document intelligence technical tips and the toolbox for troubleshooting docktail as I mentioned my name is lo Sanchez and I'm joined today with saket do you want to introduce yourself hello everyone my name is Sak Bala I'm part of AI solution success team we help implementing AI products with early adop do captured any learnings and we improve uh those learnings by improving the product thank you and so some of the content today will be presented by saket so just a quick overview of how we want to structure today I've been talking for a little bit I'm going to talk for about three more minutes just to give you an overview of what we going to cover today but then we're going to spend the rest of the session really in the tool showing you some of uh the exting exciting things about this product and we take q&as along the the session so if you have anything again Ask in the Q&A panel and we are here to answer you so today's sessions is about technical tips and troubleshooting of document intelligence and there are some questions that sometimes you might have on the document intelligent product and the common ones are for example my values are not written in my target record or my task shows fail to process the task I don't really know what that means am I supposed to troubleshoot that the task is simply not processing what's happening how can I troubleshoot that I've trained and created my model in my subpro instance now I move it to my prod instance how can I do that so I can migrate my models between instances also simply just basically when is the training happening when I'm using document intelligence so those other questions that we are going to cover today uh we can cover that in any specific order so if you have any that's top of Mind any pressing questions just again let us know in the chat and we can take that question right away uh but if not I'll just go through through them quickly and so the first thing I want to address is when is the training happening so really just to give you an overview of the process with doel so you when you are in recommendation or autoi mode mode you upload a document to either a record or directly a document task and that's a trigger to trigger the the prediction and that's a process that is Sy synchronous in that process values are extracted from your documents and then values are and then predictions are made then those predictions are shown to the user in that document intelligence workspace and when the user is selecting the values from the document it's it's actually doing validation on this prediction and that's when you see the different confidence level really that when uh the model is being trained so once the task is is submitted here after it's been validated there is a something happening in the background that we don't really see and that's the training of the AI model and so that that's to answer when is the training happening and so that's happening in the background and that's a happening kind like at the same time as your values are actually extracted and then copied back into to the record so that's training and when you get to straight through processing then you see that because we don't do validation anymore there is no training happening anymore so when you have a model that is um enable for straight through processing you applo the document and add and it triggers the extraction again still as synchronous the values are predicted and if the threshold is met then the values are are extracted directly so there is no user validation so the there is no AI training happening say was just to answer how is the training happening and then now we're going to talk more about troubleshooting and I like to call that the toolbox for troubleshooting uh because to troubleshoot we're going to access some of the the tables in which that um the product is buil done and so it's not really always straightforward so I just wanted to put that on the slide when you troubleshoot uh first you want to be able to access the the list of your use cases uh in the list view the reason I'm mentioning that is that if you are using document intelligence admin which is a great uh experience on top of document intelligence to set up your models that's great but then you don't might not be have access to the list of use cases so to to do that you just have to type the name of the table which is diore Task definition and then a do list to to access your list and on that view what you want to do is to add there are four columns that you want to add they are called the ml solution definition for X Y and Z I'll show you that in a second but you want to add those columns to your view because that's helpful for troubleshooting and then the other table that's extremely helpful to troubleshoot is called the ml Solutions and to access that it's ml solution. list and um when you open a record on that table uh there's a related link here that says show training progress so we'll see that in a second that's uh something I wanted to point out all right so with that being said let's uh go in um the tool and look at how can do that uh so I'm seeing again I'm showing my task use case here so a task definition to get to that view this is where I type my di task definition. list and so that way I can access the use cases on the list view instead of the the the admin experience view which again if you're configuring for the first time use the admin experience but if you are troubleshooting and you're more advanced user I I recommend using that you here I'm GNA sorry I'm GNA go back to how it looks okay so that's how you that's what you get when the default list View and so I recommend that you click on that icon here and on that list of available columns you want to look at ml solution definitions for and you have OC R PDF prediction and training and so you can add that to your view and the reason you want to do that is because on a case like that one so you see that migrated use case here I noticed that I don't have anything in those and that is most likely the result of a bad migration and so if I try to process tasks on that use case nothing is ever going to happen so I straight away something is going wrong here so that's how I can troubleshoot that now if I do have a value and I again open my case sometimes it happens that I process my task and then they get back with the error and most of the time it just says fail to process this task I don't have much information here so in order to troubleshoot that that's when I then go to the ml solution table and to access that I'll type ml solution do list and I like to add the created colum because it's not always there and then if you have a most likely need to create you to add a filter on create on today so you don't see everything that's happening but um yeah so something happened here I tried to process the task it said configurational network error and that's the reason why my document task failed to process so that gives me additional context to trouble shoot so I can open that task and then as I mentioned there's that related related link and that's what I want to look at and so for that one I open that and the error here is that my shared service worker user does not exist or is not active so what does that mean or if I get to my user list actually have a favorite here so uh let's use that it's just I'm on the CIS user list and no magic here and I'm looking for the name of that shared worker user and I just noticed there is nothing in that table so here what happens is that my user is is this is a user that's in being used by the system to to trigger those workflows and this processing and it's missing here so that the system the system doesn't work here but if I go to a different instance switch to a different instance and in that one okay I see I have my my shared service. worker and it's active so confidence that in that other instance something is less more likely to happen to work so step one make sure the share worker shared service worker user is there um and then so um in that other instance now and I'm again going to back to my list of task and I see something's happening my tasks are also failing in that instance so I want to troubleshoot that so I'm going to go to the list of ml Solutions again there a training request time out and that one I don't even get the link here so what's happening is that my solution is even um breaking even before it started so in that case I think it has to do with connecting to the infrastructure to do the training and uh to troubleshoot that our go to my list of system properties so I type system properties. list and I look for the glide. shared service. URL uh property and look for the value by the way this is obviously recorded but this is also documented in the FAQ okay so you don't have to memorize everything today but um if you want to go back and do the steps later you can refer to that document right now I just wanted to show you so that system property here I find it and then I look at the value and see I have a problem in here so something happened that the value got changed and it's missing something or the the URL was not the right one so if I save that I'm just going to give you the link to the article there you go so I updated that value and if I now get back to my list of tasks if I create a new task now click on process and uh a good way to see if some if the now the training is happening properly is to go again to that list of ml Solutions and look at what's happening here and I see that it created a new record just now and the record state for now is waiting for training and we'll wait a little bit but if that doesn't say that the network or connection or any error it means the system is ready to process that task and then uh I fixed my issue there there we go so training request was received and then it's going to send the data to the infrastructure and do all the training and predictions and everything like that so it looks like we fixed the issue there all right we good there and then um the last thing I want to show is when we have an other common issue is that uh when we process okay yeah I have the example here so I created a new table in which I'm storing my values and um I process the task and I did the extraction but when I look at my source record here see I'm I don't get anything don't get the value that I'm expecting in the name here so I want to troubleshoot that what's going on and to do that I'll go back to my to my use case or task definition and I make sure that the integration setup is is properly done so I did create an extract integration it did create a flow let me open the flow actually I open it because it takes some time to load so I have some that flow here I did activate it I have to make sure it's activated and I also in the properties make sure that it's running either as a system user or as a user with uh enough uh permission to write on that table uh so that extraction process is set up properly my task is being extracted uh again if I look at the extracted values I see the the value here is populated in the name but it's not uh being um copied over to the source record and so in that case I'm I'm going to go back to where how my table was built and I'm going to make sure that my table is all in need and and typically here in that case something like I don't have access to create and update the table so this is what I'm going to save that and I'll see on that task if that worked so click on the showing ducktail I do my extraction I submit my task and I go back to that view I refresh my list make sure that the value here is um extracted and then I open my task and now the value was copied over so that's uh that's much better this is trouble shoot and it worked now so as always uh here I check that um the the extraction the the integration was set up I checked that the flow that sparring that is activated and everything works and then I check the permission on my table and this is where the my issue was and it's working now all right uh that's it for an overview of the travel shooting I'm going to hand it over to saket now it's going to show us how to migrate the M between different instances right thank you L let me share my screen here all right so traditionally we are used to move code from one instance to the other using update sets uh in case of Doc Intel what happens is even when you create an update set your flow designer that you create as part of your task definition your uh use case and your keys and Fields these all get captured in your update set but the train model does not in order to move your train model what you have to do is you'll have to run certain code in the background script so this is the article where the script is uh written so we have to copy up this script what I'm going to do now is I'm going to copy the script go to my background scripts and paste it over here so once we paste it the first line of the script uh it asks for a CIS ID of the use case in order to get the CIS ID you can if you're using docal admin you can type in document intelligence click on use cases open the use case that you want to migrate sis ID is usually the last uh parameter in your url so you can copy the parameter from here and paste it over here if you're not using the tradition uh if you're not using the U documental admin and if you're using the native UI you can type in diore task definition. list this will give you the list of use cases you can also right click on the desired use case and say copy S ID that way you can access the sis ID of the use case once you get this CIS ID of the use case all you have to do is make sure that you're in the scope of global and run the script now things to be aware uh of while running these script now this can take up to uh couple of minutes or even few minutes depending on the number of tasks that you have in your use case and depending on the U model and one more thing is this script can only be ran using the background scripts you cannot apply this on a fixed script or a UI action currently due to the scope restrictions so once the script uh is completed you can navigate to system update sets and local update sets once you're here uh just give me one second I'm going to refresh my instance here so as you can see we see that there are four new updat sets that got created and two of them are in global so these are the parent batch uh update sets so you need not worry about the scoped ones you can look at the parent ones which has Global in them so let's open the first one which has uh just the name of the use case along with the global update set once you open you have the related link to export this as a batch XML click on the export update set as a batch XML which will download the update set file onto your local in local system along with the global update set you can see that there is one mode that says Task 1 to 50 so open on this update set to and click on export update set to batch XML once your update sets are downloaded into your instance you can navigate to your other other environment so let for the demo purpose I'm going to show you on another instance I'm going to go my to my retrieved update sets and follow the traditional way of uh uploading the update sets these are the same steps where we go into the related links and import the update set so I'm going to click on import update set from XML I'm going to select my update set and say upload and I'm going to do the same thing over here and say upload and search for the ones that we have just uploaded over here you can again go to your uh Global updat set first click on preview update set batch once the preview is complete click on Commit update set and this will complete the first Global update set and then let's go back to the retrieved update sets and commit the one for the task we will perform the same step we'll preview the update set once the update set is previewed click on Commit update set once you have completed loading and committing your update sets you can now navigate to your document intelligence use cases and you should be able to see your purchase order extraction use case the train model moved also keep in mind that there may be few restrictions on your import uh properties where f for the file limits let's say if your file exit it's uh 100 MB you can update that by going into import export under system properties scroll down for the import properties under XML format you can update the value over here up to 1GB that's all I have to show thank you ly back to you thank you very much um yeah just to reemphasize that the method that you use to move a docking tail train the model between instances so that's very helpful if you've processed documents in an instance that is not your for example production instance but you don't want to lose all that training so you use that to make sure the model that you move over get captured all that train so very very helpful and I just want to also add maybe that uh if you are using uh you know your tables and and all the the additional data or the the the scope app or whatever you're using to create your workflow that's a a step that you should also add to that process here I don't think the script captured your your custom tables and things like that so just remember to to to move that the you know the the usual way with your update sets or or your scop application but other than that I I hope today was helpful for everybody um thank you for joining and uh see you uh for the next session and again thank you so much s for joining us thank you everyone