[Music] thank you foreign [Music] oh that's that's me not forgetting to put off the overlay ah I finally made it so long ah I never make the mistake I made that mistake hi everyone welcome to live coding happy hour a show brought to you by the servicenowledge developer program it's a little bit of a different show all of our audience that knows how this works it's Friday it's the end of our work week we're trying to unwind we're trying to kick up our feet relax it's been a long work week we're gonna drink we're gonna build something new and this is all about avoiding the well polished demo and we're just building we're trying to show the real development process with the errors the debugging and all of that we are in the middle of our Vancouver release so over pranav's left shoulder you'll see that little logo yeah if you ever see that logo on our videos that means we're in a season of content and right now that's the Vancouver season which is now officially in general um availability so Vancouver uh but if you are interested in knowing what else we're talking about for Vancouver make sure to check out our content calendar you go to devlink.sn Vancouver and you'll see all the different things that we're working on for uh this entire season you'll see blogs podcasts live streams videos uh basically everything that we want to be showing off to you about Vancouver uh a little bit more explanation for what we do before we jump into that content but this is live coding happy hour and so before we jump in we like to explain that we are drinking during this um sometimes people don't drink on it and that's okay we have someone that's tuning in from India right now which is exactly midnight right now so if he's not drinking that's kind of understandable uh but to introduce my drink I am drinking a Heavenly apple cider from California so that's what I'm a hard apple cider what are you drinking so I'm not well so I have some hot water with me so I'll be drinking that and it's like as I will mentioned it's like 12 in the night for me so it's better I don't drink anything before I sleep so what you are drinking like uh I'm still at coffee too so I'm just having a big cup of coffee there so I'm the only one drinking and it's not even noon for me so that's I feel weird now anyways okay uh so now that we introduce our drinks and what everyone's drinking uh let's do some introductions oh we did we skipped introductions I'm off my game hi everyone I'm Earl DK I am one of the developer Advocates here at servicenow uh been with the platform been with the company for about two years been with the platform about six years uh mostly in higher ed uh part of the part of the developer program if you're new to this we do content Community engagement uh technical evangelism um experience building all that kind of stuff anything that makes your life easier as a developer and if servicenow ever has a question about what the developers are thinking about we uh either bring you all into the the table or we try to represent you on that site so we don't we don't actually really explain what developer Advocates do very often so I thought that would be a good time to do it do you want to introduce yourself hey everyone my name is pranav and I'm also senior developer advocate here at servicenow and whatever Earl mentioned I do the same also except I was working with a customer prior to joining service now so that's pretty much about me and passing it to Lloyd hey everybody I'm loik I'm an outbound PM mainly for document intelligence but also covering some of our our other platform AI product and I've been with service now two years as a PM in two years as an implementer and I've always loved the content that our developer locates but out there it's been very helpful for for me and for a lot of customers a live coding happy hour veteran thank you for rejoining us um we always liked seeing uh our different people across the company rejoin us because that means we're building Community we're building a rapport people get used to seeing certain people and then it's easiest for them to direct know who to direct questions to which is great okay so now that we've introduced ourselves and our drinks uh before we get into it um I would love for um loic if you can to catch us up on exactly what uh document classification document intelligence all that kind of stuff what catch us up to what that is before we jump into pranav's uh trying to build something yeah absolutely absolutely I think today we're going to talk mainly about one of our main uh enhancements for Vancouver on document intelligence and that's document classifier so we've done a few sessions previously on document extraction which was a powerful way to use AI to extract values from documents directly on the now platform so we can use the values to power workflows we're going to update tables and things like that and uh in Vancouver we are using this new feature called document classifier and it's essentially using AI to categorize incoming documents because another part of document workflows is that there is a lot of effort that goes into identifying what document is is is is what so if I get a set of documents I have to open them look at them and then manually categorize them and that's what we're addressing with document classifier it's we're going to see a very similar to extraction in the sense that I can build through the interface I can train my model and start classifying my my documents and then I'm training an AI model to do that now a few examples would be defining different categories for documents so if for example if I have a a survey a certificate and a contract and I'm able to fit that to the tool and gonna be able to reconnect recognize that and and then direct that to the right workflow or to the right um a person to fulfill the request and and then such so very powerful addition to our set of document AI capabilities nice cool thank you for the the summary and catching us up all right so now that we're all caught up right off what are we building today all right so thanks loik for telling us what exactly classification is doing so I thought of like creating a kind of a fun use case we have or at least I have in my childhood played with Pokemon cards and also like Dragon Ball Z cards and all these card games and all especially the WWE cards were really popular in India so I thought like why not I just combined this thing with our document classification so I have bunch of cards with me I have taken some screenshots from Google and all Pokemon cards WWE as well as Dragon Ball Z cards so now what I'm gonna do is like I'm gonna put this into our classifier and train it and then I'll see whether the our machine learning will be able to discover like classify each card or not so that's kind of a use case and definitely we invite loic always so that he can help me out while I'm building because if you have watched the last stream the Bible document document classification and sorry document extraction so like was like okay go here go here and not so he always helps me out so yeah going to my screen now let's start building this thing it would be kind of a similar experience so let me share my screen I hope you are able to see my screen right yep I can say zoom in yep zoom in all right so first thing is like we'll go into our document intelligence product and you can see like now if you have played with document extraction before so you might be familiar with this document extraction the admin experience and everything but now you can see a new kind of a new like modules in our application that is document classification that's there so what we have to do is like as always we have to create a use case and all so let me go ahead and create a use case so and once you create a use case we'll provide it with certain kind of data like the Pokemon cards or all these things I'm going to provide it and then we're gonna see whether it's like whether it's able to classify certain things so firstly we're going to click on new and I'm gonna put life coding happy hour as a title and just document classification and let me submit it so once I submit this thing and let me open this thing up now what we have to do is like we have to create categories because we have right now I have three types of category I have the Pokemon cards I have the WWE cards I also have a Dragon Ball Z cards so I'm gonna go into these Field section and I think like this might be confusing for the developers who are watching us why we are creating certain Fields so maybe in future release we might rename this thing up but it's more of like category so I can say just Pokemon category and I'll just say live coding happy hour I'm just giving some names and I'm just creating three categories as of now so let's go and create and I done with the two category and let's move into the last category that's a little bit enough that it was mainly W it was still WWF when it was when I was a child yes and I I was also watching WWF only like it was like back in 2000 it was WWF then it moved into WWE or some yeah that's there I hope it's WWE now I haven't watched anything yeah all right so now so we have created these three categories right now Pokemon WWE and all now it's more of like creating document task so document task is something like we are providing some data to our uh to our model so that it can train so let's go ahead and create first entry so I'm gonna say Pro key cards Pokemon on cards that's I'm that's gonna import and now what I'm gonna do is like I'm gonna attach some Pokemon cards let me go into my training data and you can see that I have all these cards with me right okay Pokemon cards and I'm gonna just upload this thing so it's like I think it might take a minute or two to upload because it's like a 10 MB file but similar kind of process we're gonna perform for WWE cards as well as we're gonna perform the same kind of setup for our Dragon Ball Z cards and then once we have these things then we're gonna train these cards and all uh from our ml model I'm gonna show you how the training happens and all so let's give it a minute for to put it in so till time it's uploading I think I if I can open a new tab and start creating multiple entries I can I don't know whether I can do it or not but I can try so like I'm trying to do multitasking side by side maybe I can remove this all I hope it doesn't crash let's let me do another one you gotta do the um Regular window and Incognito trick oh yeah I should have opened it up but yeah if it's synchronous then it's gonna just infinite load for you yeah that's there I can do it so I can share do that maybe or we can just you know reminisce about days gone and having the time of playing Pokemon cards and it's it's done now we have two more tiles that we have to create so that's there till once we submit this thing up Let it submit and you can see there is a kind of a new entry that got created with all the attachment that's there and similar thing like now what I can do is like I can just open this thing up and now for the first time because the machine learning doesn't know what type of car it is right so I can just I have to go ahead and classify these cards first time manually so that we are telling our ml model that these are the Pokemon cards so I'm going to do Pokemon all of them are Pokemon cards and I I don't know like how many of you have collected these cards in the past but I have done that I have spent a lot of money on these cards but yeah I don't regret it to it oh these are pretty high quality uh images yeah these are like I just got it from their official website for all these things let me just yeah we have to maybe we can just do Pokemon cards I'm just zooming it a bit I don't I know it's like so maybe this is something like you can't scroll in it if it's via zoom in we have to zoom out so that's the hip is that true you can't scroll down yeah I'm unable to scroll down that's what I realize just now while playing with it so I had to zoom out like the window and that's kind of like tricky yeah you can see that all right so where is it called all right let me reset it and now I'm gonna submit it up once I do the submission you can see that the status will change to done and this field thing will be populated with Pokemon thing Pokemon category all right make sure we Zoom back in yep thank you all right so let's let it save it takes a minute or two or like but running slow today for me but yeah for live coding happy hour yep exactly and coming back to like what I was saying I've spent a lot of money on Pokemon cards like in the past at one point I was like what I'm doing but I didn't stop I still spent a lot then my parents were like why you are buying all these stuff yeah I feel like a tradition for live coding happy hour that every developer Advocate eventually build an app based off of something they've collected because I've definitely done the same for board games I think Brad might have done the same too and now you're doing Pokemon cards which is on it's Tradition at this point so it's okay this is our guilty pleasure uh collections yeah and I think like I was a really big fan like because I was I think of five year old when Pokemon was released and uh it was just super exciting and like after a few years there was another show that came up like it was like Beyblade or something so it was like we had to collect all these things so we had to we we went out like bought those Beyblades and everything so yeah that was fun time and now let's go here okay I think I have created two use cases already let me just I'll open the document classification what was that warning so warning was like it was because when we opened it in the doc doc intel only so doc Intel haven't found like it's still running the ml model in the back end to do the extraction thing that's where that warning came up so for the first time you can ignore it oh so I think like it doesn't like more than 10 attachments so it got failed so I think we have to create another one I think what I'm going to do is like I'm going to put WWE cards to I'm being creative with the names today so I'm just gonna put like one less and like I would like to ask you why there is a kind of a limit for per document task to 10 or something because I uploaded 11 and it wrote an error that we can't do more than 10 or something yeah I was gonna say I've never seen that error and I'm wondering if it's something on on the platform side I don't know if there is a system property that maybe controls the number of attachment on a regular task but it's not something I have seen and I don't think it's coming from our product so something maybe on the platform gotcha so it's it says like maximum number of attachment on a task exceeded so maybe you're right there might be a property on that so but yeah I just reduced one image and uh and last one we're gonna do DB red cards so so I'm just uploading right now all the files uh just providing it with the data and I have this evz card also so I'm just putting all these things I think I spoke too soon I think I found the property and I think it is on the Intel side [Laughter] but it looks like an easy system property to change if if needed gotcha yeah I think I've already done that we only have three use cases so that's fine so yeah so this is for classification right so it's it's not like you need to have a million examples so that's probably defaulting to 10 may be your the product telling us like don't worry so much about providing too many examples yeah I think like the Earl you're right about it maybe because uh what it is trying to do is like I think there is a I was reading some documentation we need to provide at least 50 entries like 50 docs or something so that it can predict well it's more of like per task we can upload only 10. that's kind of a limitation maybe like once I go ahead and train these things me that error will come up so because I had a trained model for everyone if this model doesn't work because I wanted to show you guys that yeah we can classify these things so uh we now let's open the like we're gonna do the show in dock until this is the error that Earl you were talking about it's more of like document intelligence is trying to go ahead and open the like trying to extract data from it so it's running some manual stuff in the back end so that's why oh it's running the ml model just to extract the data right now we are doing classification so it's we can ignore it so we can just put WWE all these things and I'll Zoom again because you can't scroll zoom out and oh it reminds me of playing with the universal recorder in RPA and move like this so do you know do you know why there's no scroll yeah yes it's a UI bug it was fixed in the version that was released uh two days ago but uh yeah yeah it was a little bugger oh I didn't updated the latest release so yeah folks make sure you have the latest one otherwise you that's don't make the same mistake that I did so all right folks so this is something that uh we have done keep going keep zooming all right so the status is done and the field is set to WWE Live coding happy hour I don't know like why I added that live Shooting happy hour but yeah that's fine so now last thing that's left is because you can see if I refresh it so we have two which are in status of done we have the last one that's a DBZ card so let's do it for that also and I can say urgent I can do that all of these are Dragon Ball Z card um Mark is asking if it's two days ago was that do you mean GA or Islam yeah it's actually a store app so it's a plug-in and we release a bit more often than the actual that's right Comedy release so it's a version 3.1 that was released on on the 20th uh along with Vancouver GA but in that case it's just the store app all right thanks thanks for telling me yeah I saw that there was a release on 21st September right and uh yeah I didn't updated it because I was like yeah I'll update it later and I saw this bug so yeah it's fixed for people who will try make sure you have the which version it was like I think 3.1 right yes that's the one right all right perfect so you can see now I have all the three document tasks updated now what I'm gonna do is like I'm gonna click on train use case so once I click on train use case it gonna take all these three in the WWE DBZ and Pokemon cards and try to run an ml model you can see right I'll Zoom it a bit again so you can see the limit is 50. like it's saying that if you want more accurate result you should use like almost 50 uh documents and all so I just had 10 so the result might fluctuate but we can see like if it is able to classify at least I'm expecting at least 40 percent it will be able to based on name and model and now like for people like what like who wants to know what's going on how much what's the progress of the model so you can go to ml underscore solution dot list the stable ml solution table you can see that this is the new model that's been created and it's training you click on the show training progress once you are here you will be able to see the model is running so just it I think it takes like two three minutes uh and because if you only have like 25 or 26 uh screenshots so it might take like even like two three minutes that's all so once this is done then what we're gonna do is like we have some un like images that I didn't use for training so we're gonna put those things in a document task again and try to like see if our dog it's whether it's able to classify it or not so that's there so yeah let me just we're gonna wait for a minute and what I'm gonna do is like I'll just side by side create some document task to save time because uploading takes a while I know my current happy hour is supposed to be like the real Dev process and pranav is trying to multitask to not wait for loading bars but I wonder how people would feel if like we did two real Dev process and anytime there's a loading bar I just like look to my other monitor and just go through messages and wait I want to put on a show or catch up on a home I'll get a football match or something that's what I'm trying to do I won't save it I'll just upload it certain things so we have the training data and these are the like data that I didn't used right so two of each let me upload it and till the time I'm gonna go back and yeah we have the progress okay so those weren't those weren't part of your model training correct oh yeah now I'm gonna just refresh it okay so I think it's kind of run attaching it so the other screen won't load so that's fine but yeah we have set it maybe like we're gonna wait for a couple of more minutes until the time I'd like to ask the crowd like who are watching us live or if you're watching like uh like if you're watching after the live show how many of you have collected the cards and if you are okay sharing how much money you have spent if you remember so that's a thing and yep let's give it a give it a minute because yeah it's uploaded cool so this is uploaded and I can click on the training bar again and I should have opened two windows like one in English like the private window and the one so I think yeah so basically you can see that it took like just two minutes to train and our training is done let me just I'm gonna go back to MNC I'm not sure we had a question from Jace um maybe you can either of you can answer it uh Jesus asking um I know well before Dawkins Hall was based off of our own data center training models uh is the document classification utilizing generative AI at all or is it still based off of our data center um doc Intel models so I think Luke you want to take that yeah sure now this is not using it's no it's no it's not using gen AI in any shape or form it's not using four hours not using an open AI API it's not using the now llm it's just using very Advanced machine learning model but it's it's not considered generative AI yeah it's our own model um it's our own uh machine learning AIS that we've been that's part of dock Intel right yeah so to answer your question Jace you were way off base there just gonna be Chase all right so I think like uh what I so this is the files these are these are the untrained images that I have uploaded and uh what I'm now I have clicked on the process task like this is similar to how uh basically we we are just processing it and again like based on the the MN solution that we just trained now it is using it and uh trying to find like the classes and all so we're gonna wait for this particular training to happen I think it might take less than one minute yeah it's all done so if I click on now show in dock Intel and you can see let the card load now and we're gonna see each and every card to just to see whether it was able to classify all these things yeah from the name I know it's a Dragon Ball Z card and uh so yeah let it load up it takes I think two three minutes to it just just out of curiosity what's the size of each file that you have there just 40 megapixels each file is only two terabytes it's okay all right so this is there and so this is again I I like I don't know like this this is hilarious like so that was my first reaction too right when you showed the images like oh wow that's very high quality it looks like oh that's why it's loading slowly and it's like there I think there is some bugs in the UI that's for sure because when I click on it it just click goes to the next one but we are interested more of like in okay in the first one so that's there it's like Goku's image so it should be Dragon Ball Z so you can see that it's coming as 52 percent based on just like seven or eight images that I uploaded it's able to say like okay it's 100 PBZ like the second one is like the Moltres that's a Pokemon so you can see that it's saying 45 percent it's a Pokemon card and Onyx is like again a Pokemon card so it's again saying 46 percent so so the good thing about that I liked about this thing is like it's able to classify these things like if you just provide like five ten images so it's almost like score is plus 40. we have for WWE cards it's like 37 percent yeah and last lastly we have like 60 percent for WWE card so yeah maybe like they know John Cena more so John Cena and I think we have to zoom out and we have the last one that's again A Dragon Ball Z card so let's see it's able to say it's 51 percent so do that can I ask um do we know what um how it's working under the hood for this like is it interpreting the entire image together or is it looking for like only the words um is it considering layout of words and stuff like that what what's exactly going into helping it figure out which is which so I think like loic you have to take this thing I had kind of like I did some experiment but I'll share later like you can yeah it starts using a little bit of both uh it's using a little bit of the layout like shape and um yeah kind of layout of the the image and then it's also extracting whatever text is possible to extract and using that as well kind of combining both the text and layout aspects of it to to generate the prediction all right interesting and that's something my girl or I also played around with it right so you can you might have seen a trained model that I had so if I had something like a different color card for example like for the Pokemon one we I uploaded like almost the background is similar but if it was like a blue background or a background change so this score was reduced it was like coming around 40 or 37 like that so it it takes like the color the background and everything based on like my playing around with it all right so that's there and I think it's able to like based on the results we are able to see all these things so one thing I would like to try is is more of like a look like I would like to ask you is it possible like like if we can just Auto set this thing up rather than doing manually so I have couple of more cards that I can utilize and maybe like is it something like can we if I click on this enable straight through processing so it will set the field automatically it will categorize automatically is it what is it how it works yeah yeah that would work all right so do you remember it's like extraction for autofill you would still have to open the UI but the value would be there as autofill and then if you wanted to if you if it was good enough to not have to do the validation that's uh for full automation AKA straight through processing and that if that when that model is trained enough then you don't even have to do validation so that's the same as extraction gotcha so do I need to set any kind of autofill threshold or fully automate threshold because like I'm not sure how to because we are just getting like plus 40 percent let's let's test it whether it will be able to do that so I have like I can just go ahead and click on new and uh best two when did you sorry did you set up something I missed if you did no I I didn't set up anything I just went ahead and enabled that enables it through processing that's it the discretion is one so it's a hundred percent accuracy so oh confidence level so that it won't trigger but if you put something like 40 oh 0.4 0.4 .3 if it's 30 confident it would bypass manual validation yeah I think you like 0.4 probably is okay because I think a lot of the ones you had earlier was enough yeah like a lot of your correct ones were like 50 40 I think 0.4 will be fine 0.4 I've added and uh like what is this Auto fin threshold I'm not sure about it maybe maybe it's like deal in the document like uh the admin one this particular experience it should fill automatically here yes so if you look at that example if you autofill threshold is 40 40 percent and you enable autofill when you open the UI that first value that was 45 would be selected and pre-populated in your field that's the threshold here so I think uh let me just create click on new button and uh probably I'm gonna because I didn't process the old data so I'm gonna use the same thing and I'm just putting one of each just uh we could do one at a time actually so that we don't have to wait for your um seven hard drives of pictures to load yeah I think I've already started it so I just put three so it's like three thick 10B like if maybe in any second it will be done I hope maybe in 10 seconds or 15 maybe yeah not so yeah that's there and there's two and now if I do something like process tasks it should automatically like classify those things uh and and just to clarify certain things to our audience so right now the the new data that I'm putting up here so that won't be considered like into your the solution that you have already trained until unless you train it again right so the new data the test one and test you that I had so that is something like uh this data won't be considered in your solution model until unless you just click on the train button so that's an FIA for you guys process yes and saying that it's processed okay and I have to click on document tell show talk and tell and let's see if it is able to correctly classify so yeah the first image is Goku one so it is able to categorize that DBZ Coco automatically the second one is the Pokemon one and the third one is for it's a WWE one yeah don't don't worry about this image there is kind of a bug in the UI so it's loading up yeah that's there yeah coming up so yeah I think so that's pretty much it that I wanted to show in the demo and I'll also put this kind of all these cards in a Google Drive for you folks if you want to try it I don't know whether it's a it's not available on PDI but if you have any kind of a demo instance or something if you want to play around you can definitely play around I can share all these things up so I'm gonna go back to laughs okay so all right I mean we're this is all about the classification of documents right so after that you would also train models for the actual interpreting the content within each document um but the important like let's talk about use cases of why it's what what would document classification would lend into like what is there so I think the invoices like if you talk about the actual business use cases we it's we used for invoices like if you have invoice something or you have certificates and all if you like for example showing me the other day like you have all these certificates that you get from maybe now learning or other places whatever the certificate you do if you upload this thing as a task it can automatically detect okay this is a certificate this is an invoice and then it's easy so that there is no manual intervention that's there and like probably you can add on to it like what are other things that other types of use skills that we can utilize a document classification no the thing that's that's uh that's pretty good uh I think broadly they are mainly two uh categories of use cases one would be when we had to classify a large amount of data because uh they are they came in like an uh unstructured way like we get a lot of certificates and invoices and they need to be dispatched to different teams that could be used like that uh it could also as you alluded to uh it could be used to feed to one of the extraction models so if we have a model that's able to extract invoice it might be different than our model that can extract purchase order and so we want to be able to classify the file before filling it to to a model and then I would say maybe like the third example is when we are uh documents that are the same use case but very different layout or template an example of that is sometimes invoices it might be uh cases where you have a specific model for a specific type one type of device in that case you can also classify different invoices based on their layout and then feed that to different models because that's how you get the best performance that's interesting um that one of the other things you mentioned in there this is actually uh Deepak um ask like oh the simple task of classifying unwanted images like an email signature which comes as an attachment in email so I can think of the use case of hey if you have when you want to do this process Summit is your invoice and your purchase order and this and a third document and document classification you can put all those attachments in one spot and then it will classify them and then you can run different processes uh based off of those documentations but also if you if it doesn't notice what that document classification is it just doesn't go out so that would automatically exclude things like email signatures or social media icons and stuff like that because that's not part that's not going to hit any classification or you can make it classify it as unnecessary things that don't process basically so good good question and it I mean there's other ways to also address in the platform how to get rid of those kind of things via size image size or um but different ways to do that kind of thing but uh technically speaking yes document classification could probably handle that too and we would probably say like there's a lower non-data Center things to do to be able to do that but hey automate everything I guess and I was reading on like some of the dark documentation about document classification and all so like for example I had these different types of card so each particular category can trigger a different workflow or a flow like that that can be attached to a particular invoice for example if it's an invoice you can trigger a kind of a create a ticket or something for invoice related or trigger a particular flow around it so that's kind of power it is maybe that's a use case for the next for next time uh we can just try to run some workflows around it or see how it works I I always say workflows so I'm not saying workflow workflow it's more of a flows so keep in mind of that I mean we're talking about Pokemon cards and dragon and Dragon Ball Z might as well talk about workflow and throw it back to everything back then yes all right so um nobody in the chat was talking about their collections or anything like that and I we've briefly talked about it like did you want to share about anything that you might have collected in your time or as a kid or even now it was mainly Dragon Ball Z card I grew up in France in France it was I don't know why it was mainly Dragon Ball Z and then Yu-Gi-Oh or something like that uh fun fun story my dad at work at like a printer that was able to oversize something so I got like a small Goku call and made it like big but then plastic wrap and everybody at school very hard so that's awesome oh man that's so I um I said this in chat but I I only collected Pokemon cards when I was a um younger but then when I got out of it and went on with my life I still kept all my binders of them and left them and kept him at home because I was like oh maybe one day they'll be worth something or I can give them to my kids or something way back then then I went to college and then I found out one time when I came home that my mom had given them all away and I was like oh my childhood it's all gone and so whoever whoever ended up with all my Pokemon cards out there I hope I hope they brought you joy and happiness which is funny because all of my nephews now um like entertainment it seems cyclical because like my the things I was having fun with in like high school and college and stuff like that is all the same things that all my nephews are into now like Pokemon Star Wars uh Minecraft like they're all playing the same things that I was playing back then too so it's kind of like hold on to your stuff you never know who and Dragon Ball Z is like Timeless everyone's still like Dragon Ball Z I just finished Dragon Ball super so it's like okay just yeah I thought of wearing a Dragon Ball Z t-shirt for this show but I thought like we had the Unicorn one and I think like I it's like my funny story like when I was a kid I was collecting WWE cards also and I think they released only 150 cards in India I had 148. so I spent like maybe like almost a thousand dollars to get last two cards that's how much I spent and my now I see those cards I'm like what I'm gonna do with these cards I spent so much on this maybe a thousand might be a lot but maybe 800 at least that's how how much I spent just to get last two cards in the collection no shame whatever no shame don't worry about it I mean okay I was talking about board games earlier I have like almost 200 board games and a lot of them if you think of them as like 40 to 50 each you can make your own calculations in your head from there yeah anyways we we also collect shame extreme games also now I don't know we found out um recently that a number of us in our the community team there's like three or four of us that have hundreds of steam video games and our libraries are just met I thought I had the largest Library out of everybody I knew but our community team leader has like doubled the amount I have and I was blown away I was like what and then we then we realized he also worked for a gaming company before and developers would give him games so that made sense but still hey if you have any of you play video games out there contact me in front of you probably would be interested about chatting about it yeah I have like 100 as of now in my esteem and that's not even close to what a land Mark has so maybe I'm playing the small leagues as of now yeah I feel bad for calling out mark on this but yeah I have I just looked at my profile I have more than 800 so this is it's about it's an 18 year old account so yeah it's overtime a lot of it was like bundles okay don't shame me all right cool um we had um honey is talking about more use cases so like uh he's mentioning hey well in in HR you can segregate segregate documents as personal experience uh pay slips basically anything that might be dumped into a single like upload there's a great opportunity to divide them up for the different processes and why you would want to divvy them up before you get to like another Dock and tell um process or different flows is because building you want things to be reusable and you don't want to over complicate your processes so if you're building another model that is trying to deter not just determine Fields but fields for different types of documents all in one model or a flow that has nine if statements first just to determine what the document is you still have to figure out how you get them to determine which document that is so you always want to like separate your processes and then document classification is going to help you get to the point where they're you're taking away that manual step of making sure your user is putting it in the right spot which you know users never put things in the right spot at least my experience um or having someone sitting there putting them all in the right spot afterwards so that solves that and that's good with the same you get to determine the confidence levels you get to determine how manual versus automatic that process is um you get to choose are you proactive about classification and doing it manually or are you reactive and if it gets sorted um in a different place then you get to move it and you get to retrain it so giving you the tools to figure all those things out is always a benefit for admins and people that are getting these processes done both on the end user side and for the agent side so that's that's why these tools are so amazing to see because it makes jobs easier cool and that being said uh we're approaching the end of the show um we let's do announcements and then we'll talk about our drinks so um we are still in the middle of Vancouver like I said if you are interested in seeing what we're doing go to devlin.com Vancouver next week is a bunch of different things um on this channel you'll be seeing multiple streams about hacktoberfest which starts in October uh if you don't know what hacktoberfest is I genuinely encourage you to go check out um uh I had a link for that devilink.sn slash October Fest to know more about it um we'll have a live stream for next experience hacktoberfest on Thursday and then General Octoberfest on Friday next week um you'll we have some fun um stuff coming out next week about it you'll see these things popping up here's our like the Oktoberfest logo here is the crew that will be working on it in the background so we have a lot of fun stuff coming out and engagement materials check that out uh do you have any announcements or anything you want to share I know you have a link actually like that you wanted to talk about so so um go ahead and plug that too just mentioning the usual that we run the AI academies and more broadly platform academies uh it's a content for the community and uh you know it's very helpful so I recommend everybody checking that SM dot Works AI Academy and I like to plug in folks like build with RPA challenges going on so make sure you go ahead and conquer the build with RP challenge sorry about my voice it's a bit low I've been speaking and for 30 minutes more than 30 minutes so yeah go ahead and check out build with RPA build with RPA it's a contest going on right now get it the the shirt that pronov's wearing you can win you can earn that shirt so go check it out all right so that's announcements let's talk about our drinks um we usually if you're new to this we usually grade these from a scale from one to five in increments of 0.25 if we were successful which I think we are usually bump it up a 0.25 just because that means our dreams helped us in some way maybe who cares um I was drinking the Hamley apple cider it's actually very tasty I would probably give this a solid four yeah coming to me I was supposed to drink my hot water but I didn't got time to drink it so yeah I'll still give it five coming back to Louis I did finish my coffee it was pretty good still 4.5 nice okay uh if you have any questions um you can find us on LinkedIn um or message uh pranav or myself to get in touch with any of the people that you see on our shows and then we'll send you the right way if you have any questions specifically about document classification you can always come back to this video jump in the comments and we try to answer questions as we see them or send them off to the right person otherwise just contact us directly we're easy to find and then we'll send you on the right way thanks everybody hope you have a great weekend happy Friday um I'm gonna stop drinking it because it's it's only noon so I'm not gonna finish my drink anymore uh but thanks for joining us have a great day thanks everyone thank you [Music] thank you [Music]