Building Apps in the AI Era: Technical Demo

Building Apps in the AI Era: Technical Demo

Show Video

hello everyone welcome to another innovators  live and another beautiful cloudy day I'm   Debbie Cabrera your host for today and we also  have a wonderful guest for today's session where   we where we'll cover building applications in the  AI era before I introduce my guest let me review   a few things here first and foremost our chat  if you're joining us live you are able to ask   questions and share comments in our chat and one  of our moderators will do their best to answer   them and we'll even select some to answer live  at the end of the show if you're not joining us   live the chat feature isn't available but you can  see the questions that were asked and answered no   matter how you're joining us today make sure you  go to cloudon a.wig google.com and you'll see all   of our upcoming live episodes and all of our ond  demand sessions as well okay let's get on to the   exciting stuff Our Guest today is going to take  us on a journey through a fun and educational AI   presentation and demo bologi rajam welcome bologi  it's always a pleasure to see and to work with you   how are you doing doing well thanks for having me  I think this is my fourth time and I'm supposed   to get some gift as part of this yes you are a  veteran a veteran innovators live um so thank you   so much for being here uh you always have really  insightful information so I'm not just saying this   I do love having you on here so can you give us  the scoop on what you're going to share with us   today yeah we will uh um talk through what are  some of the things which we are hearing from   our customers and uh we will talk about how things  have changed since like 2023 which was the year of   massive gen hype but now that is starting to land  and people are starting to really look at how can   I use this to improve my business value uh to my  end customers so we will go through some of the   key points which we are hearing from customers and  then we will do a fun demo where we build a recipe   app uh using our gen tools which it's not meant  to be a very complex app but it really illustrates   what are the tools and services which Google cloud  has for you to quickly iterate and build a pretty   decent app yeah and I love this demo we've talked  about this we love food right we we're both big   food fans so this demo is actually really helpful  because it helps you look through all the recipes   that you might have uh lying around yeah and if  you have any questions about the recipe there's   a chat bot where you can ask questions and we will  show all of that later today and uh for people who   have registered ahead of time you probably will  have access to a quick laabs where you can it has   a it is a built-in environment where you can try  out this entire lab start to finish without having   to set up any of your environment so uh please  check that out and for people who are joining us   later I think we will have a Cod laabs Link in the  resources tab where you can go and try out this   lab there are detailed instructions for you to try  out yes exactly and we'll go through how to access   that quick laab in a slide so don't worry  about that uh but for now do you want to   get us started let's get started so these were  typical yesterday's applications say you are   a family traveling to Greece and you're looking  for kid-friendly hotels in Greece you go to your   typical website and search for it and it throws  up a whole bunch of results and you are left to   figure out what exactly is a best match for you  so but model applications a lot more interactive   right it asks you questions tries to understand  what your need is and really gives you the results   which you're looking for so that you're not uh  left with the task of figuring out things for   yourself right it is much more intuitive and  personalized and that's the direction which   we are seeing and even now uh the expectations of  consumers have changed where we if we go to a chat   website like a e-commerce site we expect be able  to ask questions and find out more information   so when you to start building those modern  applications these require an extensive array   of capabilities to be successful first is scale  previously your applications were restricted uh   most of the time catering to like a very small  audience which is based in your geography and   so on but that has changed right but with the  power of social media and so on you are able   to reach consumers far from across the globe  so you application needs to be able to scale   to serve millions of users so the other one  is uh security right which is becoming very   critical because there is always some amount  of uh security threats going on at any point   in time so your applications you build have to be  reliable secure and continuously available that's   also a huge thing because if you are running  an e-commerce site you cannot afford to have   like a 8 hour down downtime right uh one of the  customers whom I have worked with in the past said   one hour of downtime cost them a million dollars  they are a fortune 10 customer but it may not be   that as big for other customers but it's still  important and then you need to have the ability to   innovate right provide new and U Better customer  experiences to your end consumers so that you can   differentiate yourself from your competition uh  so this uh the platform which you use needs needs   to give you the ability to experiment quickly and  move faster and ultimately lot of this is built on   data foundations where you have to have ai ready  databases that you can use to actually power your   application because gone on the days of using  Legacy databases like to power your applications   you need databases which can do a pleora of things  so like when you start building applications like   that in the past that's not going to work in the  modern era and when people who have worked in like   traditional Enterprise applications you have the  vpf apps and you have the vpf databases so the   vpf apps focuses on developer toil agility cloud  observability and security and the VP databases   focuses on database performance scalability uh  security of the data U making sure their data   platform is future proof and you have advanced  capabilities for building gen applications   previously they used to a lot of them used to  operate in silos so there used to be this whole um   U dance which needs to happen to coordinate with  the database team to ensure that you can deliver   an application on time and within budget but  now they need to be lot more integrated together   so that they can work off of each other to build  the kind of services which the business needs yeah   and just on that I feel like I say this a lot  maybe if you heard me before I'll sound like a   broken record but yeah data and ai go hand in  hand right you're you might have great AI but   if you don't have a database that can power it if  you don't have good data it's not going to be as   great as it can be really yeah and um ultimately  when you Leverage The Power of AI the data is   your secret Source because you if you have been  a business who have been around for a long time   you have so much data which you can leverage to  personalize the AI Services people who have used   gen tools know that yes gen tools provide some  certain answers but when you personalize that   with your own data and ground that in your own  data that's when the true power of AI is unlocked   yeah absolutely cool so when it comes to um  building modern applications if you look at uh the   operational complexity which uh Enterprises deal  with that is is actually hindering the in ability   to innovate and compete there is so much uh time  wasted between Contex switching when it comes to   uh developers they are in their documentation they  are in their IDE and so on so that limits the uh   productivity gains which you can have and uh when  you look at um um the duplicated effort to build   and deploy applications which happens across lot  of industries that means that you have a slower   time to Market and then you have technical debt  and Legacy databases which really hinder your   ability to innovate and build more modern Services  uh if you're using a very Legacy database you can   it cannot be used as a vector store uh for your  AI applications that makes it more costly to   maintain and it really limits Innovation uh  and then lack of standardization also means   that it your applications could run into high  security and compliance risk as well especially   in regulated Industries like financial services  and Healthcare this becomes a huge challenge so we want to make sure that the technical  teams have the best experience to build a   future basically adopting AI supported development  practices uh for building your applications so   Google cloud is your partner for building modern  applications we have infrastructure which is built   for scale right our Google kubernetes engine or  gke can uh operate at 10x the larger scale than   any of our competitors right um and when you look  at AI expertise and Leadership ship this whole   wave of J was triggered by the Transformer paper  which we published in 2017 and and we have been   using AI for such a long time it almost becomes  kind of second nature to I think we published a   paper in 2016 where we talked about how we reduced  our power consumption and our data center by   50% using AI right and people people one of the  things which people don't see is how we use AI to   actually filter out all the spam which you used to  get in Gmail I think it is like a 3 billion pieces   of spam messages which we filter out which you  don't even ever see so that's how powerful and   our use of AI has been for the longest time within  Google and we also have industry-leading databases   where you can have a globally available uh  database like spanner which is consistent across   the globe uh and available all the time right and  we also have alloy DB which is the Improvement   upon the postris engine which can function both  as your transactional data store your vector   data store and and it can even do some amount of  analytics and we have worldclass security right   since 2020 Google cloud has had 75% fewer uh  critical and high city security issues than our   competitors yeah few things on here it says uh you  know Google uses AI to write more than a quarter   of its new code yes this this strikes me because  recently I I haven't been able to get my own   statistic yet but I feel like it's been helping  me write more than a quarter of my own new code as   well yep have you experienced that oh uh very much  so because uh one thing which I tell people always   is when back in the day when I learned computer  science I spent a lot of time learning syntax and   semantics I think that is actually going away you  no longer have to worry about oh is there a colon   here or is there a semicolon here because that was  a language semantics now you can ask code assist   for help and it actually completes that for you so  that is essentially getting eliminated and it lets   you focus on the more creative aspect of coding  yeah and you mentioned alloy DB we recently did   a road show in Q4 and we did a few alloy DB Labs  those are also available on code labs and I just   want to if you want to play around with that you  can really see how you can do kind of everything   in one place absolutely um so one of the areas  which has been a lot of interest to customers   because when you start building applications for  the more modern era you need to have a platform   which can actually cater to all the needs of your  applications and when you are a large Enterprise   you need a platform like that uh so that uh  different teams within your organizations   are have consistent way of building applications  that makes it more maintainable ensures it follows   all the security and compliance policies of your  organization so and so on so we work very closely   with saber uh saber if you make any kind of  Airline booking I think it pract practically goes   through saber I think 99% of the time um so we  worked with saber to actually help them build out   their application platform and we have had a very  strong opinion as far as when it comes to platform   engineering is concerned and uh another example is  a customer called regology where we work with uh   them to actually help them adopt alloy DB actually  just like what you said uh because they felt   like that is a much better uh way to leverage  their data to uh uh improve their chatbot and   uh provide um services to their customers and the  next one was we helped uh we were working closely   with tuning and there have been an early partner  for code assist right from uh it launched nearly   I think well over a year ago uh at this point uh  and they saw uh productivity gains of around 33%   uh when they started using uh um um code assist  and that means that they have boosted developer   productivity faster time to Market and much more  streamlined on boarding yeah again here with the   productivity gains I just feel that a lot in my  own experience I was just uh creating an app last   week and I had set the whole day to be able to do  it and I actually had some free time after that's   because you're smarter no it's because of the  Cod assist yeah truly it's it's really changing   the productivity gains game true and millions of  developers are building with Gemini right Gemini   is providing so vital wide variety of services  uh to actually build new applications especially   with Gemini code assist we have seen so much  productivity impact internally there's a 40%   faster completion time for a common uh Dev task  50% less time writing tests 55% less time writing   new code because the code is actually getting  generated for you and I think one of my favorite   things is actually the 55% less time understanding  new code AS developers you you are always thrown   into fixing or updating somebody else's code and  uh back in the day when I used to be a developer   um writing code every day used to spend so much  time trying to do a dry run of the code trying to   understand what exactly a block of code is doing  and now you can just write select the code right   click on it and say hey Gemini explain the code  to me yeah and also another one that I love is   that you can use Gemini to comment the code as  well so not only do you make it easier for your   yourself but also for that other person you have  the comments in there and you have the explain   this function exactly for a lazy dollar person  like me who don't write enough comments that is   a fantastic feature with that I'll hand it over to  Gemini uh to uh Debbie to talk not Gemini Debbie   to talk about what's next you can call me Gemini  if you want no no kidding kidding of course okay   so what's next and that means what's next after  this event first you can continue exploring this   lab so once this live broadcast concludes today  you'll have the opportunity to explore the lab   in another platform as we uh bology mentioned  before Cod labs and you'll receive instructions   on how to do that in a follow-up email so  don't worry if you aren't able to do it now   or you haven't registered early what else you  can try out more Google tools so you can join   our innovators community and develop new skills  with 35 no cost monthly credits to use on courses   and Hands-On labs in Google Cloud skills boost  which is the lab that we'll be uh using today   in the live demo or will be the lab that you get  access to today and then number three join us at   Google Cloud next 2025 in Las vas discover new  and upcoming Innovations at Google Cloud next   and you can get inspired by the technical talks  we have you can tackle School skill boosting   challenges and get your questions answered as well  as connect with other fellow developers and we're   really excited to offer you the lowest ticket  price at $999 if you register by February 28th   using the code Dev 999 VIP at checkout to redeem  this offer I'm really excited about next this year   very much so do you remember the penalty kick  demo which was there on the floor last year yes   which was fascinating to me that was a cool use  of AI uh tools yeah yeah and I know some I know   some other people that are building some cool  demos uh for this year as well so yeah and the   exciting thing about coming to next is hearing  from all the other people who are using Google   cloud in so many innovative ways and you hear  about use cases and scenarios which you normally   wouldn't 100% getting in front of people that  you don't normally see every day it really is   the best Okay so let's get into accessing the  lab so what I like about this lab is it covers   a lot of the points that you made right bology  about modern data development data foundations   and how you can Infuse AI into that and kind  of make them go hand inand um and that it uses   a lot of different products right we'll use a  Vertex a agent builder of course Gemini code and   Cloud assist uh big query cloud storage collab  Enterprise notebooks and more so we're going to   get a good view of all the different products many  different products and services that you can use   in the cloud uh I really love this demo okay but  before we jump in let's go over how to access the   lab so if you're joining us live live go ahead and  open up an incognito or Anonymous window in your   browser and then you can go to the quick laabs URL  we have on the screen here explore. quickls dcom once you're there you can sign in with your  account now make sure it's the same account   that you use to register for this live webinar  your email was previously registered so if you   try to sign in with a different email it will  not work just make sure you're using that same   email and if you're just joining us and you didn't  register earlier you may not have access to this   lab but the good news is there will be a very  similar version of this lab as we mentioned   that you can complete in COD Labs on demand once  once this event is over and again you'll receive   that information via email so once you've signed  in you can go ahead and click the Box building   apps in the AI era technical demo it'll look like  this image here and from there you'll be taken to   the classroom to start the lab now very important  don't click Start the lab until you're ready to   start the lab because there is a time limit on  these labs and once the time is up you will not   be able to pick up where you left off you have to  start from the beginning these do have to be done   in one shot the good news is that this version  of the lab will be available until the end of the   California day which is where we are today 11:59  p.m. Pacific Standard time so you have all morning   afternoon or evening depending on where you're  joining from to complete that y yeah so if you   want to do the lab after today as I mentioned  there's that on demand version and you'll get   all those instructions okay and if you're having  any trouble reach out via chat to see if we can   help you get access all right let's go on to the  demo thanks debie and uh at this point uh um one   thing which I want to rrate is two things one is  um making sure you are in an incognitive window   we use Quick laabs all the time at our various  events and that inevitably happens where people   open it in a window where they're already signed  on to with a different set of credentials and it   creates issues and the other thing which I want  to say is I think the lab is for 4 hours or yes   4 hours but um we did it in events and I think  most people were able to complete it in an hour   or an hour and a half I think yeah maybe yeah  it probably won't take longer than two hours   uh but just in case there has been times where I  have been like oh I'll be right back and then we   yeah don't start it when you're in the middle  of a three-hour meeting yeah yes cool so the   first thing which we want to show is around  Cloud workstations so Cloud workstations is a   environment in the cloud I love this so very much  because it lets me have different workstations for   different demos and stuff which I do so I have  one for this I have one for the different set   of demos I do so I don't have to Tinker with the  environments each time I need to do a demo and   the nice thing about this is uh it work you can  actually use it with your um local ID as well you   can connect from your local ID into your Cloud  workstation and use it as if the entire thing is   on your local machine but from a organization  perspective it makes sure that all the files   and the data stays in the cloud right so once you  here uh uh the pre-warming step of the quick laabs   should already have created the cluster and have  the uh workstation ready for you all you do is if   it's not started click Start it'll take couple of  minutes to start up and then you click launch and   then you will get a window like this which kind  of looks like um the opening of it's actually a   managed vs code right vs code instance and you  will download the code to your machine and uh   this is actually what it looks like so it's a  very simple application right it's a recipe uh   search and a cooking advice application fun fun  stuff um and uh we will go into the uh uh code a   little bit later but one of the first things which  you do as part of this is actually creating um   a data store for an agent agents are all the rage  these days uh so we will actually start creating   an agent so we have something called an agent  Builder and so obviously since this is a demo   I have done all of those steps yes just like the  cooking shows right uh so we have data stores here   in one of the data stores we are actually what we  are doing is we actually just storing files so if   you go to recipe uh old cookbooks folder and if  you click on one of these files this is actually   The Complete Book of cheese oh wow my favorite  book yes uh Complete Book of cheese and we have   a whole bunch of um cookbooks which are in the  public domain which we downloaded and uploaded   to a Google storage bucket and these are old  cookbooks right if I'm not mistaken yes yeah   these are all in the public domain so we don't  want to viate any copyright so absolutely yeah   so we uploaded all of this to our Google storage  bucket and here uh when you do create an agent   you can actually specify a data store which can be  like a bigquery database um a cloud SQL database   it can be a storage bucket as well so what happens  when we do this is it actually takes all this data   and it's almost creating like a inbuilt rag  like a retrieval augmented generation so it   indexes all that data and does it for you so that  way you don't have to deal with the task of oh I   need to find the right algorithm to do the in the  create the embeddings I need to find the right um   algorithm to find the uh do the searching and so  on so all of this is taken care of you taken care   for you because that makes it much more easier  for you to build an application so once the data   store is created you actually go back and you can  actually create an app so in this particular case   we have done a recipe search app which actually uh  sorry this is actually the The cookbook search app   app where you can actually go and search and the  fun thing is you can actually search and test it   here too before you even build app but we have  actually built an app which we have deployed on   cloud run already so in this case we can actually  search recipe chili recipe oh chili delicious   it's cold so I thought we might use chili so  one of the things which we have done in this   particular search is we have actually uh grounded  it using Google search as well so it is actually   pulling stuff from um the both the uh books which  we have have and also other places as well so if   you look at these are all the results which it  found in the QuickBooks which we uploaded so it   has things mothers used to make a collection of  oldtime recipes right so so this is a quick way   to create a agent and to actually test it out as  well once you uh done with that you can actually   go in and create another data store I think  that will be one of the steps in your uh this   where you actually create a data store against a  bigquery one so for the bigquery one one of the   things which we are doing is we have a spreadsheet  uh uh with lot of uh information about recipes and   that we are going to use our collab notebook to  actually upload to bigquery and this is actually   one of my I am not a very smart person when it  comes to uh collab uh I am pretty dumb I know   how to be dangerous you're inexperienced let's  say that yes maybe that maybe that you you put   it much more nicer than I did right so what it  does is it actually taking a CSV file which has   all the information and you can actually use um uh  uh collab Enterprise to actually consume that data   and upload it to bigquery so what we are doing  is we are taking the uh information parsing it   retrieving the right information and pushing it to  bigquery so one of the things which it can do is   it can actually suggest code as well so actually  you start doing it as you can see it types ahead   and gives you suggested code so so code assist  actually works within your collab Enterprise as   well so if you're somebody who is new to collab  Enterprise the easy way for you to get started is   just write a comment in natural language and  collab Enterprise will generate the code for   you yeah and I love the you know doing it this way  because you said like you can play around with it   first you can test different things you can use  Gemini cusis to adjust the code if you want you   don't have to go and change it in your app right  away you can kind of do it in here and see how it   works you can ask it in natural language to remove  comments uh sorry remove columns rename those   columns add things so in the lab you'll see that  as well you kind of test out all the different   things that it can help you with along the way  um and make this even faster yeah and it has an   underlying uh runtime engine which can actually  run this code and give you live results as well   so once you do that you have pushed all the data  into um the the um the bigquery instance which we   are using and that is actually what we going to  connect uh using agent Builder into um into the   uh data store yeah and I will say I've used agent  Builder uh a few different times at this point   probably like seven to 10 different times and not  only is it constantly improving I feel like each   time I go in there I see something new but it is  really quick like in a couple clicks you do have   a search app or a chatbot or something like that  spun up to kind of try out things it's very very   fast yeah the first time I tried it I'm like I  I don't understand anything about gen let me try   what this thing is and I was was very surprised  with the results which it was able to get yeah   um so uh this is the end result of actually  processing the uh spreadsheet data using um   collab Enterprise and uploading it to um bigquery  so you see all the data which has been upload   loaded already this is a schema for the database  table in bigquery where we have the directions the   title and the URL and you see all the individual  documents here and after that we actually built   an application one was for uh searching through  the cookbooks which we showed you earlier and   same thing for a recipe search as well so um both  those applications have been created now uh let's   go to the actual code actually uh it's a very  simple uh streamlet application um somebody I   don't know much about streamlet somebody showed it  to me it was a very simple way to create a quick   web application so it was very helpful yeah I  actually used it that app I was telling you about   that I was creating last week I needed to put a  quick front end on it so easy and simple just pull   it in make a few edits it really is great yeah so  this is actually uh some of the code which you see   let's ask what is this code doing so I'm going to  use Gemini CODIS to say explain this to me so let   me in this part oh yes this is a great thing to  point out since this is a separate uh area like   if you're using a Gemini Cod assist anywhere else  you might have to log in here and just make sure   you're in the right project maybe enable an API  uh because we want to make sure that the right   project is logged in for COD assist as well as for  whatever else you're working on so you will find   this uh every time I'm doing a lab or something  I find myself yes you have to log in a few times   just to make sure that you can use it in the right  place cool and so now Gemini code assist has gone   ahead and explained the code to me so it's saying  initializes a chat session with a large language   model and configures to use specific data store  for grounding its responses and it's also breaking   down the individual steps of what it is doing  so it's a very easy way for you to understand   what the code is doing and make changes to  it much more easily uh So within your code   please try this out I know it is tempting to just  copy paste the commands and just rush through the   Cod Labs you have four hours to do the code  Labs so please make sure you take the time to   try out all the things which we are mentioning  in the quick laabs as well as the code Labs uh   so that you can actually understand the power  of these tools yeah especially if you're used   to programming in maybe a different language than  what some of these uh Labs have or you know what   this app is in uh then you can really understand  what's happening by getting the full explanation   and it's very thorough as he said it goes through  every single line and tells you what's going on   exactly and one of the other things which I  used to like to do in code assist is uh can   you create a test plan for this because I used  to write hate writing test plans oh yes testing   very important very important yeah of course  we all do it yes must we don't we don't want   to test information yeah some people do it but  I prefer not to so it's actually going to create   a test plan and give you instructions on what  should be covered as part of the test and um uh   let's see so uh what it's actually using um unit  test mo. patch to MOX or Vex AI in it uh and it  

is actually providing you all the steps which is  necessary to create a test plan for this right um   you don't even have to leave the ID right so you  have your code there you're like oh let me I need   some test can you just help me with that and also  if you have any other questions about the other   files that you have open or other Google Cloud  questions you can ask how to do it right here in   the Gemini cusis chat in the ID I think that's a  great point because very often all of us used to   have like stack Overflow open Google Search open  documentation open while we writing code this   actually lets you stay just in this window and  ask the appropriate questions to move ahead with   that yeah you know we all have well maybe not all  of us us some of us have many many tabs open all   the time and once you go out into tabl land it's  hard to come back sometimes how are you doing on   time jimy I think we're good on time okay yeah we  have 30 more minutes wonderful so let's look at U   the the code where we actually calling the uh uh  the app so in this particular case we are actually   making a call to Gemini to actually retrieve  the appropriate data when uh there's a chat   message is uh sent and let's look at the other  recipe search one which is also doing something   similar where we actually going to um the let's  actually do an explain this on this so live demo   I love trying out different things I feel like  he asked me how we were on time so he could show   this special piece here which is exciting explain  this file and I'm not joking when I say that I use   this all the time I know it sounds corny like oh  they're paying her to say this no no really I do   I use it all the time it's just very helpful  because even working at Google Cloud I mean you've   been here what 10 years eight years eight years  yeah my me five even all that time there's so   many products so many services so much different  knowledge that it's nice to have that friend there   that you can kind of just ask different questions  I know bology and I both tend to work late at   night you know so when you can't uh you know  ask other people those questions because they're   sleeping Gemini Cod assist is there to help you if  then you can't phone a friend you ask Gemini right   yeah so uh so I did uh explain this file in this  and it's Gemini has come out with explanation of   the entire file on all those steps which is doing  the import steps the search sample step where it   is actually uh retrieving information from our  data store and it's also telling you how the   streamlit app is functioning right whether you  when you the query block where it is actually   get retrieving the results and displaying it back  to you and we actually did deployed this on um   Google Cloud run uh so let's go to um the the  agent builder stuff where we uh actually we go   let's go to Cloud run yeah where we actually have  this application deployed so Cloud run uh have you   used Cloud run before Deb yes I use it a lot I  love deploying my apps there I feel like it's   the easy fast way to deploy exactly because it is  completely serverless you don't need to create a   cluster or really understand all the tuning which  you need to do to make an application function   well so you can easily deploy it and the nice  thing is it only really consumes resources when   there are actually requests coming into it right  so this is a recipe app which we have deployed   and let's go into that app and here's a URL for  the app which has been deployed and and what you   mentioned Cloud run right so when you're not using  it you don't have to worry about undeploying it or   anything like that it only uses resources when  you're you're calling it when you're opening the   service got it so we have our recipe search app  in here and we have our cooking advice app in   here so let me ask the quicking advice app on uh  for some recommendations uh while that's coming   up oh yeah I wonder what what should I have  for dinner today this evening cooking advice   I'm going to ask like I have some be friends who  are vegan oh okay so let's see what can I use as substitute only if I can type better maybe J I  can help with that actually a quick note on that   sometimes if you misspell things uh not in your  code but if you misspell things when reaching   to your AI uh it'll actually still understand  you which is very nice maybe you've used other   llms and chat Bots that are out there and you  notice this but it really makes it so that in   natural language you know not all the time do  we type correctly or say things in a way that   is completely comprehensible you know so so I'm  going to pretend I made that spelling mistake   on purpose oh pretend it was totally not so  I'm asking hey what can I use as a substitute   for butter to make a recipe vegan so it gave me a  whole bunch of options like applesauce uh neutral   oils coconut oil peanut butter almond butter all  those are options oh that's great and and honestly   I have a lot of dairyfree uh friends so you know  making brownies or anything like that applesauce I   hadn't considered that fantastic nice cool let's  look at recipe search uh let's say I have a lot   of mushrooms in my house mushrooms yeah not not  the magic kind mushrooms just right um mushroom recipes oh mushroom migas I've never heard of  a mushroom migas me neither I love migas though   well this is so this is great because you  know when I want to make dinner and I say   okay I have one leak and some turkey you know  what can I make with this um this is wonderful   right and we created this in what I mean in total  it'll be like an hour an hour and a half at most   yeah and you just have an app that can help  you with everyday tasks of course but also   in your business with other with other tasks  aside from everyday uh home tasks fantastic   so that pretty much concludes our demo awesome  yeah and great demo yeah nice that's uh is there   anything else which we should show well no I think  actually we're going to go to some questions but   I do I just want to say that I love this demo uh  I mentioned because it touches so many different   Services right we went to agent Builder we had  big query we created the data store uh then we   used Cloud run and it was so quick and it shows  kind of like the sheer difference between how   much faster we can develop applications and  software now because of these AI resources   that we have and how seamlessly they integrate  into our current development life cycle right   we've both talked about how much that's changed  for us uh so now we're going to take a quick   break for questions and we'll select a few to  answer live on air and we'll see you in a sec and we are back okay let's get into answering  these questions let's get right into it so   a question came in which tools to choose When  developing an AI app it the the broad answer is   it depends right It ultimately uh depends on the  use case which you're trying to service because   with developing there's no AI app per se right  AI app for me means like the Gemini app right   more about like apps which are powered using AI  I think that's the way I think about it I think   that's probably what the person who asking the  question is intending so depends on the use case   whether if you're using say U the great example  is what we showed in the demo today and which you   can try out on the quick laabs where how quickly  you're able to leverage AI within an app right uh   because the if you look at the overall code which  we wrote the part about utilizing AI is probably   like two or three lines at most rest of us was  a standard app all we did was made the call to   the right API and retriev the right information  I think that's where it boils down to and if you   are looking at more complex applications where you  need to chain results from one Ai call to another   and make decisions based on that then you start  looking at tools like Lang chain and whatnot yeah   and you know to go along with that you can decide  on do you want to Lo use a proprietary model an   open source model an open model there are many  different uh you know ways that you can access an   llm as well so and I I think I think that's a very  important Point Debbie because one of the things   which verx AI lets you do is lets you choose your  own model because we have our foundational models   which are Gemini I think the Gemini 2.0 flash  which went GA last week I think and uh we also   have like third party models like models from  anthropic Mistral and so on available and we   also have a whole variety of models from hugging  phas which are open source models available for   people to use so pick the best model which is  ideal for your use case I think that'll be my   advice yeah exactly and even uh the most recent  deep seek you can actually try it out on vertex   AI oh I haven't tried that out that's be fun and  I think vertex has over 300 different models you   can try so yeah perfect okay next question so why  is AI important for modern apps um more than AI   important for modern apps the way I look at it is  customer expectation has changed so drastically   right I think when we were all stuck at home for  Co our expectations of what an app should do or   what uh the kind of customer experience we hope to  give have hope to get has changed quite a bit um   and I think one of the stats which I read a long  time ago is it takes you 66 days of doing the same   thing over and over again to actually become  a lifelong habit and we were a lot of us were   stuck at home for a lot longer than that and that  has really changed our expectations out of how we   look at um uh what we need from a company or an  Enterprise and so some of those functionality for   instance being able to have an interactive  conversation about a particular product say   if you're buying say even a prepared food from a  store you want to be able to ask hey what are the   allergens in it say if you're allergic to certain  things you need to be able to really understand   how uh it works and things like say if you're  buying a piece of clothing hey everybody has a   different body type how would it look on a certain  body type being able to ask those questions and   interact with it or maybe even see that piece of  clothing on different body types I think that is   a feat one of the capabilities which we released  a long time ago um as Google and so those kind of   capabilities are becoming more and more important  into the app and I think that is why providing a   more modern experience is made easier using a  I think that's a way I think about it yeah and   and also what you were talking about before right  with the personalization so the body types or you   know we've the the sunglasses try on things right  every day that's getting better and I would also   encourage what you mentioned before like look at  your use case don't just assume like oh I need to   use AI you know I there's also an argument to say  like AI can probably be integrated into almost any   use case that you have but really don't just try  and use AI because you think you need to really   take a look at your use case and understand how  it would benefit you good okay so next question   so how would you integrate AI into an app there  there are a bunch of different ways right we   mentioned some of them but yeah you can do what  we did today you can make an agent you can call a   model into your app to uh bring in grounding from  search or rag uh you can use it to structure your   data or collect your data recently uh with a group  we were doing a little bit of a hackathon and we   built an app where we uploaded pictures of our  clothing and then we used the model to kind of   put metadata to it and say like this is a black  blazer this is a sequined silver Blazer so kind   of playing around with the multimodality of it too  I think can be really fun and you can find a lot   of different ways to integrate uh in that way yeah  and again goes back to the initial point you were   making right around what is the use case which  you're trying to solve for and that also dictates   to some extent how you integrate AI into your  app I think I think uh you were there last week   at developer week uh here in Santa Clara where  one of the demos one of our colleagues did was he   was running everything locally from the browser he  was using web Ai and uh G JMA JMA 2.0 to actually   power an application which was completely running  locally with no internet connection because he   wanted to show that as a use case and that that  and that worked really well so It ultimately   boils down to what are you trying to accomplish  as far as a use case is concerned and I think   that will determine your strategy yeah absolutely  um oh this is a good one alloy alloy DB I feel   like that's really your wheelhouse uh are there  any worthwhile performance benefits of using alloy   DB in a non-generative AI app versus just staying  with something like Cloud SQL uh uh definitely I   think one of the even though the alloy DB uh  product is built on top of a postgress engine   we have done so much work into improving the  performance of it uh I forget the exact stat but   it is so much more performant than the standard  Pro PR uh instance which you get through Cloud SQL   so and the other thing is alloy DB also provides  you the capability to do analytics so if uh if the   use case for your um application is hey I want to  I don't want to invest in a huge uh data warehouse   for my application but there are certain analytics  use cases which I would still like to use I   think alloy DB is a fantastic fit for that yeah  absolutely okay so how do you collect data for AI   models so again there's many different ways you  can do this there are you know obviously if you   have your own data that could be csvs PDFs uh uh  you know Word documents anything like that uh or   even now going into the multimodality right videos  images different things like that and uh then also   you can it it can help you structure your data so  even if you have unstructured data uh it can help   you do that as well as we have all of our public  data sets that you can access right and uh kaggle   is a fantastic source for data sets even before  gen I have downloaded different data sets from   kagle to run models against it uh I think once  I downloaded a whole bunch of x-ray images and   used our pre uh tool called autl uh to actually  see hey what can it detect what's happening in a   particular x-ray so kagle has amazing data  sources uh uh data sets which you can use   gets uh started on it yeah and we also have a lot  of public data sets in our bigquery bigy bigquery   has a lot of great public data sets Okay so what  uh what is the list of programming languages that   it supports I'm going to assume we're talking  about cist here yeah so I think there's over   20 different languages that it supports yep right  so the fun part the fun part about the answering   this question is there are 20 plus languages it  officially supports right um I remember I think it   was last year's next when somebody came up to me  and asked hey does it support power Builder I said   I've never used powerb Builder let's try and I  gave it a natural language prompt and it generated   a pretty good power Builder script and it was not  according to me that it was good because I don't   know anything about power Builder they said it is  a pretty good script so the the thing about the   llms are they trained on so much data I think the  languages which we say we officially support are   the ones we are continuously regression testing on  and we have done enough validation on but a lot of   the time I've seen it supports a whole lot more  than that yeah and just to we did get an answer   in the chat for exactly the languages so I'm just  going to read that out but it is there if you want   to look at it bash C C++ C Dart go Google SQL Java  JavaScript cotlin Lua mat lab and I'm going to   stop there but you can keep reading there's a lot  that you can play around with and like bologi said   maybe play around with something it says it might  not support and test it out kind of push limits   there okay if I have little to no experience  coding can I use these tool to build a complex   application that cages to millions of users or  will I need a team of developers also if I already   have the uiux done can I upload to the databases  files and the AI will help generate the back end   so um I remember one of our colleagues had a like  a graph of the benefits of coding assistance uh   so the benefit of coding assistance was one the  why AIS and the experience of the developer was   on the x-axis so the mapping was people with  like little to know experience coding assistant   are not that helpful if you have a reasonable  amount of knowledge and then at the other end   of the spectrum people who are like expert coders  they're also coding assistants probably more get   in the way rather than actually helping them uh  so what we have seen is yes if you have a little   bit more experience with coding that is much more  easier and coding assistance can generate code for   you but it is not a substitute for understanding  how to design good code how to structure different   code and so on so it understands it at the code  level but it doesn't understand it at overall   application Level the one of the questions was hey  if I have the database and the back end already   built can we build that application right actually  one of the things which I've seen is people take   the schema and convert it to an AP I right um um  I think one of the capabilities within um Gemini   code assist is the ability to generate open API  spec 3.0 right uh so it can you can actually give   probably give it a database schema and get it to  generate a spec and from there you could probably   say hey from this API spec generate a stub app  for it obviously it'll just be like a marked   up app you'll have to connect it to the back end  and do all those things but it is quite possible   and we have tried some of that out as well yeah  and I will say to reiterate what you were saying   the human in the loop portion is something that  we you know constantly remind people of at events   and webinars like these because um it's meant to  help you Gemini cusis is there as a helper it's   not meant to do everything for you right so you  will have to go in there and check and sometimes   you will have to make edits because you want to do  something slightly different than what it gave you   so yeah make sure you're kind of checking in uh I  don't know if it would uh yeah make sense to cater   to millions of users just off of what it gives you  especially not in production so always make sure   you you check that y uh okay so how much does AI  app development cost that is a broad question so   if you're talking about um creating a very simple  app using Gemini APS it doesn't cost that much it   is fairly straightforward to do but if you're  talking about hey I need I want to trade my   own model then costs start adding up so I think  I don't know if there's a better way to answer   this question than that yeah I would agree with  that and it also yeah it depends like how often   are you going to be calling the model right like  which model are you using there's actually pricing   online that I don't know off the top of my head  but there is pricing for how much each call input   and output is for our newer models you know I  I know definitely 1.5 and 2.0 are on there um   so I would encourage you to go check that out  yep we have a handy pricing calculator on the   web to actually try that out yeah exactly okay I  believe that is all of our questions for today yes   yeah thank you so much for engagement yeah yeah uh  we love to hear your questions so you know thank   you for being here for engaging uh any parting  thoughts apology um no the only thing which I   would say is that quick laabs is a lot of fun to  do yeah uh if you already have access to the quick   laabs try it and if you um are joining us later on  demand try out the Cod laabs yeah and if actually   I'll answer one thing here maybe some of you are  saying quick laabs code Labs what's the difference   here they're just two different platforms where  you can do the labs quick Labs kind of sets up   the environment for you you can get right in and  start doing it whereas the code lab you uh need   to do a little bit more of that leg work right so  Cod Labs will be available indefinitely and there   will be some free credits that you can do and  you'll receive that via email and the quick lab   as we mentioned will be available until 11:59 PST  so uh please please enjoy doing that thank you so   much for joining us for another innovators live  uh make sure to go to our website and check out   our past sessions and our upcoming sessions and  let us know on the web if you complete these Labs   you know baly and I would love to hear from  you how you feel about it thank you so much

2025-02-27 04:03

Show Video

Other news

Fixing the MemoryWriter 6040 floppy drive with Solvespace 2025-03-30 15:08
Bumpy Week for Nvidia, Tesla, Billion Dollar Space Bet | Bloomberg Technology 2025-03-28 17:01
WSL Setup (GPU WiFi Cracking, Wireshark, Remote Desktop and a lot more) 2025-03-28 05:42