[Music] thank you [Music] thank you foreign [Music] happy hour this is the show that you get to watch us try and figure stuff out as we figure it out it's a great time really honest it sounds like a train wreck in slow motion but it's not because you get the Insight of the discussion the learning the discovery all kinds of things let's go around and do some intros first this young lady over here hey everybody my name is Lauren McManaman I am the newest member of the devel oper advocacy and I am here to show you doc until today so I'll leave one I think it's my second attempt at running the show so uh wish me luck and I'll toss things down below hi everyone I am Earl Duque I am a developer advocate here at servicenow I've been with the team for about a year now which makes me the second oldest developer present developer Advocate on this is just three of us right now um uh but before that I was with higher ed with a bunch of customers doing development and bring it back over to Chuck my name is Chuck tabassi senior developer Advocate I got more time than both of these put together so it was serviced now since mid-2010 was a customer for a couple of years before that and have development history going back about 40 years so yeah I'm the old man of the crowd so get off my internet you kids I'll get off your internet lawn let's go this way to uh describe what's happening with the beverages today again I'm being boring so last time I I said I was doing so for October this time I'm also doing sober October so I'm just doing a ginger ale today what about you Earl not that darn it that was really good I had a long week so I'm drinking I have a golden state cider I drink this um about a month ago on air so um I need to get rid of some of my cans so I'm drinking this cider again from uh Sonoma County California where I live and it's a Jamaica Flavor flowery sweet hi Micah is that Jamaica in Spanish I mean how Michael the flower [Laughter] and I am having high quality H2O from my developer MVP 2021 canteen instead that's about my fourth or fifth one of the day very nice with the Beverages and introductions out of the way what do you say we talk about what we came here to talk about so Lauren you are in charge today I am part of the the whole uh Tokyo program Pro features that have come out with Tokyo and one of the things that was definitely on a lot of people's lists is the concept of document intelligence so document intelligence is part of I don't know what branch it's technically under but essentially it's giving service now the ability to read on our behalf so part of automation engine automation automation engine thank you so under the branch of automation engine we have dock Intel so basically we're giving servicenow the ability to read in PDFs um and basically we are going to train them to either put those things directly into some sort of table or transform that data before we dump it maybe using something like flow designer so essentially uh that's what we're focusing on here today uh use case wise uh I've selfishly I am going off of a use case I am kind of tinkering with myself uh and so I figured that uh now the world is opening up there's a lot more travel happening I'm having to travel a little bit for work upcoming uh we're doing like a little bit of a meet-up tour uh and so I thought maybe creating a travel app would be kind of fun um with a lot of these uh OCR Solutions like I I think it's interesting to utilize documents that don't always look identical as well uh you know obviously we could do this with just like a standard PDF that always looks the same um however that's not really the most thing like relevant thing uh with businesses oftentimes people will scan stuff and it'll look funky it'll be a little bit maybe out of line um people might type things up in their own fonts and so instead of doing what I initially intended with it being um uh just like a flight itinerary I thought it kind of fun to do passports uh passports or something that's commonly scanned in if you're managing them uh so they get a high variability of looking different uh while still relying on a similar format we're only doing American passports but that's what we're gonna be focusing on here today so I'm actually going to go ahead and share my screen uh I'm going to be using my secondary monitor as long as it's uh it's been performing a little aberrantly recently so if I have to switch things around in the middle of the show I apologize but for now it seems to be holding steady so I'm going to go ahead and share my screen number two fantastic okay so document intelligence uh this is if you were wondering where this exists on your platform um and if you're not seeing it that is because it is a plug-in that needs to be turned on so if you have access to this based off of your uh like obviously based off of what you've purchased on the platform I'll always add that asterisk um this is a plug-in that has to be um kicked off from your plugin manager first but once you do that is it plugged in or is it it's the store I believe uh I I turned it on Via a plug-in there might be multiple ways to do it um but I I did it through a plug-in a couple of days ago yeah I believe it's a licensed product as well so yeah you mentioned U.S passports only and I love this use case because uh you know it is one of a kind of document that can come in a number of formats I from a business perspective you might think of getting invoices from One customer has a certain format getting invoices from another customer has a different format so yeah you could take this what Lauren's doing here today and adapt it quite easily to you know European passport or Australian passport or Japanese passport because they're going to have different similar things in different places absolutely so um that's kind of what we're gonna be tinkering I'm playing around a little bit with today um we can always try using a different passport I only downloaded um a bunch of fake US passports well actually the ironic thing is I don't know how fake some of them are uh obviously finding demo data for passports is a little a little goofy um some of them obviously look like rather fake uh however someone else I found seemed to be like examples of people that got scammed from like like romance scams which is a little goofy I did get a Canadian password to test out um like a fake you know Michelle Obama Frank Moss whoever that is so we've got some some demo data to uh to work off of Today We Salute You Frank Mott yes we appreciate you and your gullibleness with romance scams for that we have examples today to use I I want to start a thread of a slack channel to find out where people get their demo data because that one just intrigues me usually I go to like mockeroo.com to get a Json object but no Lauren goes to fake romance scam sites to pick up passports well technically I went to Google images and they were there from Google images and Pinterest some people were just putting their password on Pinterest which God knows is probably not the that I sure hope they're fake because that is not a good observation of pii yeah I got no I approve me if they're fake or not so fingers crossed so this dashboard so obviously with all of our products we try to give reportability on how they are working and these kind of give a hint of what makes up doc intelligence you'll see things like transactions and tasks and task definitions which represents kind of the main artifacts that represent doc Intel so the first thing um before we create anything is we have to create a task definition a task definition on dock Intel essentially represents the use case in which we are trying to emulate so you can see here I was tinkering around with creating one for American Airlines flights confirmations but we are going to go ahead and create a new one and we're going to call this uh passport bigger font bigger fund take a drink whoa okay guy in the back of the bus you can see that on my mobile phone I think if you're gonna harass for her bigger thought I could always get bigger fun there we go so once we have created our task definition we will go ahead and click into this if it'll look for me today come on oh that's never a good sign when you get that Banner no it's not go back oh there's an old UI 11 ish screen with the Greenback girl well does it want to load okay thank you so this is what a task definition looks like essentially it represents how you will be processing all of your documents and documents come in the form of tasks so the task definition provides the framework in which all of your tasks will be processed essentially setting up like The Source table that it might be linked to um the the target of where it is going as well as the keys the keys are probably the most important piece of this because they are the fields that we are looking for so if we bring up our let's pull up Michelle I think that was the clearest looking one um hopefully this is not the real Michelle's sport but got no way of proving if it is or isn't um to use as the example for creating our keys so if we were reading this optimally like what would we essentially want to gain from this uh passport so to create a new key you're going to go ahead and click new in this upper right hand corner we're going to give it a display name and then a name itself I will say that the only two types of data that can be read by dock Intel currently are just text or I don't know why does it say string like everything else but says text um and true or false so okay so no numbers or images or anything at this point no but that is where uh this will come in handy so they can take strings and put them in in formats that are more appropriate date yeah yes dates which we will which are like the bane of my existence on Services everybody time zones so hopefully obviously these are a part of the passwords are great because they got strings they got numbers but they also have the god forbidden dates so we're gonna go ahead and just create uh let's just do one for passport number so we'll call this passport number go ahead and click submit and we're not going to create all the models create a few so while this uh chugs along at a very very leisurely Pace Chuck and Earl are there any use cases that have been brought up to you uh for like from customers or for people that are interested regarding how they are expecting to use something like dock until not for me yet I have received a direct message saying that we should use doc Intel to read beer labels while we do live pretty cute that would be more of a QR code scanning thing all right I'll do I'll just do date of birth and nationality that's all I'll do after this you can get the passport number from two places one is in that upper right of the you can't page the 910-239 blah blah blah and it's also on the bottom and that's typically What machines read is the bottom stuff correct and you'll see you want to parse that out we've got options which is kind of cool um so obviously nationality and I'll do one more I'll do a date I'll do the expiration date because that's probably the most important date on the whole thing I notice when you go to customs or something they just take that page and scan it to like almost like a barcode scanner or a magnetic stripe reader which I found out my latest passport which is all of about one and a half years old is uh has an RFID in it so they don't even need to scan that anymore yeah really oh that's nice but on global entry it makes it much easier too because all the machines are RFID now yeah I don't even need to carry my Global Entry card anymore it's like I'm so disappointed I'm so disappointed uh so now hang on let me rearrange my window whoopsies there we go uh so now that we've created at least enough keys to get the ball rolling let's go ahead and just put in one of our documents and see how this goes so essentially before you can create anything these are the first two steps that are mandatory obviously you can create flows to kind of massage the data very similar to transformation Maps um however these are the first two mandatory steps create the task definition and the keys and that you're wishing to look for do you have any info about like the stuff on top warning threshold and version uh to be honest I do not I I do not remember off the top of my head that this is a prep to get us by for now just enough to to have fun with this um I apologize yeah we never advertise this as like the expert show this is this is to play around and see if we can get it working right or at least that's what I thought it was so you're right on track cool so we're just going to call this um Michelle Obama and we're going to go ahead and attach our document so the neat thing about attachments is that you can do it in a couple of formats uh I'm gonna do this one as well should we do Frank Let's do let's do Michelle and see if this one works because I did jpegs before I have not done pngs I know PDFs work but I know jpegs work but I've not tried PNG so let's go ahead and try that and click process tasks now the second that you click process task a couple things are going to happen um essentially this not only is the attachment going to be downloaded onto the I believe server side of servicenow um but it's also going to use AI to basically scan through the document for potential data that can be stored um utilizing dock Intel so if you look down at the task itself it'll say that the stat is is processed is false we can't really do much until is processed is true it's pretty quick usually takes about a minute or so but you can also get good reporting data as far as also like on the processing as well so some of our MVPs inside the audience are actually pointing out some of the fields some of the things that we're seeing on our screen so warning threshold is um talking about these Comfort confidence scores if you don't know if you didn't know about this um the document intelligence uses a machine learning platform called nagini that runs in the background when and sends it off to a data center to process and so this is why this feature isn't on-prem yet it's all you have to have an online server but that's what that's what is happening in the background it's getting those confidence scores from an algorithm that is hosted by this Nan Genie engine as in as in Voldemort's snake yeah yeah I'm speaking of which the actor who played Hagrid passed away today so oh I'm wondering if our developers did diligence and if you hover over the field labels did they leave a hint for us of what they might be for the back on the task definition let's see any of them just hover over the label see what happens it's loading Robbie Coltrane thank you Phil so if we hover over warning threshold fail auto fill threshold fail and for um enable straight through processing I believe that is if you checked if you check that then it'll start processing the doc intelligence part without having to um press a button so if a user puts the image onto an attachment record it'll start processing it immediately if it doesn't if it already has the training done beforehand yeah if you do the target table that's that basically allows you to align Keys directly so if it's only if you've gotten this to a point where it's predicting things really really well and it doesn't need a lot of data massaging you could have it just Auto dumped directly into a specific table um we don't necessarily want to do that but you can set that up and then just align key to field within service now which is kind of yeah that kind of feels like the difference of doing a direct table API rather than going through the import set API yeah yeah exactly okay which I don't know no not necessarily um but if it was just a one-to-one for you know you could so we shall we see now that the is processed is true um so if we open up our task so remember this is processing a singular document for our use case of passports in general we can see a couple of things we can see every one of the keys that we Define so this is why it's important to define the keys ahead of time uh we can see whether or not it's actually found anything and it all appears to be empty so what's the point of having AI if I can't identify anything well point is that it has to be trained so this is the first document for a brand new task definition so obviously it's going to be pretty um zero percent accurate but we can help it along its way so excuse me if we go ahead and open up doc Intel's little window here we can coach the algorithm them on what data we are actually looking for so we have a nice window where you can you can zoom in you can kind of navigate to exactly what we're looking for here just so that everyone can see it well and on the right hand side you will see all the keys that we defined so for example if I want to show the passport number currently it doesn't have anything as a result but if I as a human being can see that the passport number is 910-239 and we see that we have something returning as a result if I hover over this it'll show in that blue box where it is reading that data and on the right hand side it shows zero percent the percent score that it is showing um is the prediction score so if this was a much more accurately trained definition it would show that as a higher degree of Competency but again first time but you know it does actually showcase that information that we are looking for so we're getting I thought you would have had to draw those rectangles you just type that in it goes I saw one correct which is pretty damn cool for the first time especially that's really cool to see so similarly as we walk down we can see there's a little blue check mark and we can see that it shows one is completed so it's the overall status completion uh of this so first name we're going to type in Michelle and again it's accurately finding all Jason sensitive goo yeah the interesting thing to me is that it's not for somehow reading the ones on the bottom I I do not know why uh it's fine uh but that is kind of wackadoo that it's for some reason only seeing because uh when I was looking at what was it flight information um my like confirmation was referenced multiple times in that document and it found all of them so maybe it's the discrepancy of fonts maybe it's that wavy pattern that is indicative of the passport I'm not quite sure but at least it did find at least one of them which is witchcraft I tell you so the next one watch it let's curious how as far as like like literally the stuff is so if I type in like Spanish words in there uh it's not seeing it's not and well it gets kind part of it like it sees that whole thing as one whole chunk if I try to put type in endorsement of course some Nails what it's yeah it's it's not quite making out the letters it's getting the Latin though is the Constitution yeah yeah it's pretty great so we want the last name though which is not endorsement it is on page 51. [Laughter] cool and then nationality so we're going to type in the United States okay this is a good example of how it can show multiple because obviously United States has written multiple times on this passport it's a U.S passport so we don't want that one we don't want this one we want the one that is right yeah you're cursoring over those gotcha well does it not see the one that we need though uh type out of America it should by process of elimination it should get that no oh interesting oh great uh oh it's grabbing that whole block oh that's not well um so this this might be an opportunity to showcase a little ellipses on here so if you have an issue uh you can flag this field so you can kind of come back to it later it'll prevent you from submitting the whole thing uh let's just go ahead and do that for now uh and then we'll finish with the expiration date so the expiration this is where you do want to draw Little Boxes by yourself I mean this is an important part of the step because this is why it doesn't work right away you need to do some training for it 10 documents to do a training model so that's this is intended part of the process so I don't know what this guy is oh exact match mode oh that's your case sensitivity so if I go ahead and try to submit it says hey there's you know something's been flagged um right now because it didn't uh identify this correctly uh I'll just say it's uh should we put it as non-applicable or missing um missing in this document yes you couldn't find what you were looking for correct so we can go ahead and click submit it'll save all of our changes you can see that Michelle Obama is now represented as done and if we close the doc Intel window and we reload the form we will see some substantial changes so not only has the uh overall status of our task been shifted to done which is good because it got all the fields that we could possibly get however it could get all of our data so if we kind of zoom in here we'll see the passport number is listed the expiration date the last name and the first name the only thing missing is that nationality um which is which is pretty pretty damn cool um going back to our past definition though you know we've now trained on just one uh passport but what happens now if we you do another one so let's do a separate one see how much better it's already gotten just based off of one uh entry type so we're gonna go ahead and click new where what's the other guy's name what's his name Frank Frank Moss all right well I I was playing rough rank early let's do Christine gionee's Christine she looks like she was arrested a splattering picture it's funny I had a similar instance happen I almost used one of my passports but it shows some you know some information that's not I didn't want out there but uh they told me I would I wasn't doing my uh passport photo taken I smiled and they went don't smile and so I think I reflexively frowned instead of just went passive and I look like a James Bond villain in like one of my pictures in 2012 they said don't smile but in 2021 they said it's okay I guess so ah So Christine is now being processed as well so hers will will flip over here in just a second if you really want to make it look authentic you stay awake for about 48 hours because anytime you traffic internationally you're going to look like you know the dog dragged you out of the dumpster anyway so yeah that way the the people at Customs go yeah that's you all right oh my gosh I had the last time I used my passport I just just while this is processing I had a scuba diving trip that got canceled so I was only in South America for about 12 hours if you ever want all the attention on you at an airport go to a foreign country and go home in 12 hours you will you will get you will get searched so many I got searched four times one time they pulled me out of line to get onto my airplane to search me again because they're like there's no way you're not smuggling stuff that's like I promise my scuba trick just got canceled I promise all right so we can now see that the field updated it shows that this has been processed so if we show this in Doc Intel we will see lovely Christine's photograph and we've already seen some substantial changes on just after one document so now if we go to passport number at a 60 chance that this is accurate it accurately predicts her passport number which is so so cool after just one document right I think that you know with machine learning algorithms and stuff what it was it was like you know so many documents to train but just after one you know it's already showing progress interesting that it missed the five on that bottom number on which bottom number go down to the second choice on there highlight it nope nope nope right there well that's bizarre oh it's business but this time it's actually reading it though yeah it missed the five but it got the U so it's like the same region offset I mean it's 40 so at least it knows it's not confident about it right yeah it's like maybe close America has to name America okay now can we get the oh I didn't find her first name and Miss Christine so if I type in Christine it finds it but it has a zero percent chance so this is obviously do the fact this is a different shaped you know document this is uh you know the crop is different the zoom is different you know it's the same document but it comes in a myriad of you know formats when people scan them in um so we can just go ahead and select Christine didn't you have one that was kind of a skew like sometimes you don't get these scanned exactly straight on I do I have one of the orthogonal in my opinion yeah I should try one sideways and see if it can do it but I'm not maybe maybe later on after it gets at least two under it's under its belt so you can see Giannis too is only a 13 I'm surprised the passport number was so strong given the other ones are like 13. but let's see if this one actually gets nationality so if we type in United States of America okay better cool that's good fantastic and extra operation date boom it got the expiration date too so out of out of the what five fields that we created after one it predicted accurately two of the fields it's not bad so this time go ahead and submit and for Christine if we go back in and refresh this guy that's running an intern we can see it's got all of our information we love our interns so now uh now that we've we've showcased that we can load our documents in both PDF and picture format all of which are legible and servicenow with a little bit of coaching can start predicting that that content I think the next step is obviously considering the imperfect state of this content right if I were loading this passport into a table and service now right having these fields in this format is not optimal um like optimally I'd like to find the user named Christine gionis based off of this first and last name uh we have a country table in service now so I wouldn't want it to just be a string of the United States of America I'd like it to be a reference possibly to the country table and then of course the the Holy Grail of the mall which is transforming a date into something that's actually a date object um disclaimer this is live Cody happy hour so don't expect miracles here don't accept Miracles so now we can start setting things up a bit differently but I need to create something else so uh similar to transforming data we need to create a staging area for these variables to be linked to um I think we can just link them to a new task in the task table so let's try that first so that way we don't have to did I understand that you could do like um I think this is where key groups come in where you can do lists of things like if I had a purchase order for example the line items could all be read in as separate records right yes so if you want to create that that was something that was experimenting with um with the uh what was it the flight information uh if you obviously you're probably don't have just one flight you're probably coming home too so like it'll show you know departing arriving time seat name whatever and then again departing arriving date sheet name whatever creating if we go back to our task definition uh that is where we can group these fields together for the possibility of an iteration so that would be very helpful for invoices as well you don't want to just create a static number of of items people can buy as much or little as they please so creating the key group uh is how you would do that so if you want to just click new I have not done this yet successfully so forgive me if I I stumble on this portion um but this is we call this the test key group I can go ahead and click save and I believe you can just add Fields or excuse me keys to that group maybe not oops specify a Target table all right to be explored that is the purpose you're going to run into situations that we have you know quantity price description amount quantity price description amount over and over again and you want to grab all of them maybe you add it on the key to the key group ah yes so you just it's a one to many okay this key is part of a larger group correct yeah like variables and variable groups if you want to think of catalog items that way yeah pretty cool alrighty going back into The Fray so if I go back in here so the next thing that we want to do is link this to I'm trying to think of the best way to do this whether the best way is to create a like what is it like a passport table for this to exist in I think that's probably the best way to do it so let's just go ahead and create the ideal passport right the idea of what this what this could be so I'm going to go ahead and go into app engine app engine Studio because that's where my app lives and just create a passport table based off the keys that we already made with their ideal formats of of real dates real sys user kind of thing yeah so I'm assuming you don't have a Michelle Obama in this user table I don't think so but maybe if she is she a uh a test user in our demo data I don't think she so we're just gonna call this the begin uh scratch uh passport Auto number sure uh don't care at the moment I'm trying to remember the flow part is the is the part I'm the weakest on because I didn't get the chance to practice all that much before but we'll we'll do our best that is the point of locally happy hour uh go ahead and add the table Phil has a good question I think we can kind of muse as we wait and figure this out so the training part of the machine learning algorithm so I know when it gets retrained or when it gets trained you'll see it getting retrained under solution underscore ml uh dot list the that table um they're in machine learning there's this weird conversation of like as you build a training model and um it gives us confidence values back and we're simply picking the biggest best confidence confidence values when we go through it who's training who kind of situation um but I think how this is working because in your they say like the model isn't official or isn't actually activated until 10 months is it activated or is it just confident confident once you do 10 tasks um through the machine learning I can't remember what they said but I but knowing that it's constantly retraining I think it's constantly watching for what we're doing so work work everything that we do over time even if we don't um go into the dock Intel interface to give it confidence about or pick values it retrains itself over time especially even when you do upgrades it automatically triggers it um so I think it it's constantly getting better so it's actually learning from us not necessarily it trying to force us to abide our process to them I don't know it's it's an interesting conversation for anything that's machine learning right um yeah I think that's what's happening in the background at least from what the demos I've I've been seeing um and there's what they're saying very cool wait why is everyone saying no in the chat oh we're joking about uh differentiating zeros from O's oh that scared the crap out he's like what did I do what did I do what did I do why is everything no I think because the um when we were showing the dock Intel learning page earlier it was uh interpreted an O as a zero now everyone's saying zero for O's instead and we're all making fun of what no could look like so there's love it no and no and no and and yeah yeah so there's lots of silly ways that we could potentially mess with the algorithm while we're doing this training I don't think I'll find any emojis on a passport say that yeah Gabby's will have one so now uh let me look at my notes here for a second just double check so I think that we can link it okay so next we create another what's our next password should we try the Canadian one probably not which probably stick with oh here's a good one with like a really weird setup though let's do oh I don't Jacqueline Jacqueline Jacqueline there we go all right and we're going to attach and then where is lovely Jacqueline and there she is she's the one that got scammed by some some fake prince online thieves and then go ahead and process tasks I believe it's at the task level that we start determining the um where this gets linked up I think okay uh no it is not at the task level well is it I'm trying to remember I'm trying to remember it's the target table or if it's the source table for the flow that I wanted to play around with I'm trying to think well worst case scenario let's just look at Jaclyn and see how that went she's not processed yet no she will be I'm curious to see because it's a pretty crappy image as far as that goes it's not the most detailed and it's got that weird format so if anything's gonna throw it for a loop it's gonna be this one all right so passport number damn still got it right though it's a less accurate percentage but it got it first name b nope not the best senior Jacqueline got that and then shiv hey got the accent on the E though that's something I wasn't expecting at the till day on at the till is that the till is that a till day or is that that's that's uh a grab accent the United States won it this one is continuously getting a little bit messed up this is also the lowest res image that we have so we'll mark that one as missing in the document and then date of expiration is accurate oh okay I was looking at her birthday yeah so not the worst considering considering the image and then go ahead and click submit so BP I'm trying to I'm getting used to reading the chat my notes and then have the demo on the other screen it's pretty admirable how well y'all do this so seamlessly Bueller Phil says he knows her no it says where'd you get that image I know her wait really no he's kidding okay I was like oh my gosh what a small world yeah trolls us as much as he can toes especially slightly nervous that gullible yeah so let's go ahead and start our flow so basically I would like to create a flow that uh retrieves the fields from our task I'm just trying to remember how I Orient that to the passport itself but starting off from flow will at least give us a good uh start foreign within AES I'm not because that's an extra step it's right there it's so close but we are going to create a new and we're just going to call this the doc Intel fraction passport extraction yeah we're gonna go ahead and click submit so this is going to be triggered off of our task so the task remember is the iteration of our use case the task definition is the use case itself and every task or every document is that iteration so that would be like if there was a new incident or a new problem that is the equivalent in this case um however no just no go away um however we don't want this to kick off the second we create our task because we have not processed the keys it does not process the document and we have not successfully approved that those keys have been retrieved appropriately so what we're going to do is we're going to do not when it's created but when this is updated um the task uh the or the dock Intel task table is di underscore task I believe the eye underscore Cask fantastic and we're going to add a condition so the condition of this will be when the status of our task is done so if we wanted to compare this to what these look like for our task um let me just pull one up as a visual example if it doesn't take too long to do so so if we go back into uh Michelle's and this is the field we're operated off of so the second that we approved and went through every one of our Fields this automatically uh shifted to done so that's what we are looking for and this is done go click save there we go so once this is done what do we have to do first well the first thing that we have to do is get our keys right we have to somehow get our keys and store them appropriately um one thing that we can do is we can look up records because all of these keys are actually stored in their own table as implied by this related list as well in fact we can see them all so if we go back to our documents intelligence application menu we can see we've got extracted values if we open that up everything that we've ever extracted from Doc Intel exists so Ergo this could easily be a lookup option so if we add an action we can go for look up records look up records you almost had it no right I think it in some cases it's one word in other cases it's two so try this one look records there we go okay look up records we're going to look up uh the extracted value table so I think it's just called extracted values or bi underscore extracted value is our table in question but of course we have to filter this down um for our specific tasks so we are going to filter this to where the task um I will say when we have to zoom in for these these shows it does get a little wonky on the flow designer it's not usually like this but it's because it's the result of zooming into like 200 percent um the task there we go is the same as our source record correct right there at the top um is this the TR but it's not right that's the test record uh hang on let me double check something I might have missed a step uh because I think we have to actually store this in a source table and that's the that's what I'm I missed in doing so if we okay so if we open up one of those what we don't have any reference we do have a reference back to the task we do okay so maybe I'm maybe I'm making this overly complicated so if we do this yeah task is Task okay let's try that let's go ahead and click save run it run it we test it out sure let's test it for Christine oh my gosh it's already 4 15. time flies when you're extracting fun amounts of data and well uh that's another wreck well no those are the records so they don't have a display while you want them they don't have enough that's fantastic that's the ones we're looking for all the ones linked to Christine okay so if you want out of those is there a value or something in there well yes so all of them uh usually in the context of these flows what we recommend is looking up the extracted values and then creating flow variables for them or something like that um or like or do something within the bounds of the loop itself um but within each uh data extracted value there is the the actual data in the data field great name for a field yeah that's what's hard too is that a lot of stuff is named task and like everything is named task and service now um so it's easy to get confused what task you're talking about but if you were making a variable you can make the variable name the key and the value the data correct key would be past partner I I gotta say I'm not a fan of that form layout what do you mean it's so easy to follow it's so descriptive with lots of help for text on on every field two wide cars two two single column displays followed by some random yeah so within the so if you didn't want to use uh you could obviously you could always use uh flow variables like that's probably the easiest way to do this um you could also use like a uh what is it a decision table so like if the key is this put the data over here if the key is that put the data over here um if the key is this first transform it and then do it um but essentially that would be the setup for any flow uh concerning uh data or document intelligence as a grand finale should we try the Canadian passport do it do it do it let's try it or take Frank and rotate it 90 degrees yeah I like Earl's comment of like task top test that's literally what half this stuff is the only thing that was more confusing was um when the new mobile app first came out and like everything was like an applets or it was just it was also like very recursive on on Navy it was very difficult um but let's go ahead and do one more task and see what happens when we throw uh old Canada in the mix for for our passports and see if we would need a new definition or if we could get by with what we have because it has the same information it's just kind of in different places how different could it handle it we will find out loading loading loading yeah my internet's a little bit on The Struggle Bus recently between like the stream and the all the windows I think it's a little bit strap for time there we go and do a new task and who is our Canadian or like a little text used by scammer it's all for like scam examples it's so funny um not it's not funny to get scam it's funny that like that's where all these passwords have come from um it's the C1 so we're gonna go ahead and call this just Canada boy that's who he is that's his legacy and we're gonna go ahead and pick his passport cool and process task you have to have the chewing song format will three of the American passports like uh would that be a hindrance or or would it trance to be to be better I'm not quite sure not quite sure if I can see this being really useful for like receipts and stuff like that if you want to create like your own we could uh create a concur competitor so that you won't have to do all the uh stuff we have to do behind the scenes for all of our cooker expenses just waiting for it to process there we go go ahead and show it in Doc Intel and see what it finds so passport number uh yeah it does Billy does not know where to look for passport number but if we give it a hint 160-379 oh no not helpful I will say the sideways text saying this is a use by scammer might be throwing everything off I didn't know maybe I mean we had a whole bunch of stuff in the background on the US ones too and it seemed to do pretty well true it does catch scam chat which had a lot which which is interesting because we're just asking about what if something is a skew yeah we bought that it caught it caught something that was a skew well it does not have passport numbers I'm missing in a document but what about his name his name is Mike his name is scam chat his name is Mike and it oh no well oh oh oh oh no oh no I'm gonna grab the sideways mic at least it is getting Mike like it is it is finding them bikes that are there but it is deferring to the ones that are sideways not the uh not the closest so we'll see what's ignore and then Perry will get Perry it get very okay nationality uh it gets like the whole block so I wonder if the resolution has to deal with it because obviously PDFs are going to be a lot more higher res than just like using a kind of a crappy image um but it does because it has a lilac like it has a likelihood of blocking in too much as a singular entity it seems because obviously we don't want all that for Canadian well what about the date so the date of expiration so 21 oh no luck with expiration date yeah this is a an interesting training stuff yeah training batch to go through we preferably you would use real documents that have some form of consistency it's true your training model can you know build off itself instead of constantly guessing it guessing wrong correct the good old challenge every data import faces of inconsistent data what do you do with it well as far as a grand finale goes I don't know how how Mike Perry treated us but it looks like we're coming up a little bit on our end of the hour correct no Kate Frank rotate Frank wait what do you mean take take Frank's picture uh passport oh open it oh yeah open it with preview just go click click and do a command r there you go hey upside down will be fine too let's try I'm gonna try oh that's you're you're I was going for 60 you went for 100. well let's try I'll see if we can do it now I'm curious so we're gonna go ahead it's very funny a little slow from my liking at the moment all right that happens with all demos you should know that you've been doing since you started well usually those are a lot more thought out ahead of time the point of this is to do stuff on the Fly in sales you gotta be refined process that's what happened for the Canada one we didn't say it right it's process task ah actually I think this um passport will be accurate for all of our Australian developers on stop oh so I have to bring his up though but I'm not going to be able to read it I won't know what I'm looking for hang on let me make this full screen and then put this side by side there we go and I'm going to rotate him back so that I can know what I'm looking for okay okay process yet or no uh the check box hasn't gone checked yet it might be freaking I was like what the heck is all this it had Australian Surfer yeah really you're a bad influence [Music] gotta see what happens with Frank oh ah there we go I'm gonna jump at the bottom of the hour this is my whoa wait wait wait wait it loaded it period it auto rotated it okay we learned something today okay that is a great finale that if it could actually auto rotate the document based off of the AI like reading the text it's like oh this is upside down that's really cool I did not know it did that fascinate well it got his passport number rights uh first name I guess Moss well they're all equal so six percent Frank surname Moss nationality boom hey you gotta write this name it's getting it right and then expiration date it got it right so the only ones that mixed up were first and last name but even with the upside down in Canada of course we're gonna give that one a break that's true well that's a way better Grand Finale than the other one easier than training an intern yeah [Laughter] too cool doesn't get me coffee though true we got a couple minutes left let's do our ratings oh beverages I mean it's hard to mess up sugar-free ginger ale um I'll give it a four so 4.5 since we had a pretty successful demo all right Earl uh I have a golden state cider uh I think the last time I gave it was before um and that's accurate it's really solid I think for us fours are actually really solid drinks on the show so yeah and I'll give my high quality h2o5 can we get drove on the air to at how what's your how you get your Coke today yeah uh if you didn't notice right above Lauren's left shoulder is a Tokyo logo so we are doing uh our Tokyo content continues in our series season of Tokyo devlink.sn slash Tokyo if you want to find all of the content that's available there you can go there by calendar by category and as of this morning Earl said it's up to date so we're keeping up to date we may be a little day or so behind sometimes uh any uh call for Content call for content is opening on October 18th which is just a few short days away from us now if you've got an idea or even if you don't have an idea we have office hours in there you can show up be part of that conversation if you have an idea we will help you refine it there are people showing up going I've got this experience and I did this and we'll help you drill into what might be interesting to an audience so you can submit a title in abstract in the call for Content when that opens up so take advantage of those office hours we are there for you and that will be running through November 15th so you have about four weeks four weeks to get all your ducks in a row get your ideas Pat it out and then we'll begin the selection process and let people know that's for a speaking opportunity at creatorcon23 in May of 2023 in Las Vegas so anything I missed uh we're in the middle of Oktoberfest still going yeah Devlin Oktoberfest if you don't know what that is it'll explain everything there uh meetups are happening next month in the um throughout uh well they're happening all the time so just check it out devlink.sn slash meetups have you been to one of your developer meetups if you haven't you probably should check it out good stuff all right that's going to take care of us for today I've got to run to another meeting real quick so thank you everybody for showing up it's been a wonderful time we will talk to you again real soon take care bye outro [Music]