A new future with AI tools by Rahat Yasir || Lightup Conference

Hello everyone. Thank you for inviting me to talk at

Light of Virtual Conference. I'm really honored and really happy to contribute to this

like Nobel effort and my today's presentation topic is a new future with AI tools. Today we'll

talk about like we are hearing a lot of things about AI, how it's like gonna rule the world

how it's going to change the world, but we will take a step back and think about all the aspects

like how can we incorporate AI in our business, what is the future is looking like, and how we

are using AI in our today's daily life. So before I jump into my presentation, I would love to give

a like a small like intro of myself. I'm Rathi Asir, lead AI developer at OCDM. It's been eight

to 10 years I'm working in this AI industry in different like like as initially I worked as a

researcher like at a like P2IRC research institute and then like last few years I'm working at

different like big organizations and taking their AI like models into production and like

productionizing them so that people can use them like those AI model or research and at the same

time I'm a Microsoft MVP in AI. It's been like six years

I am Microsoft most valuable professional and back in 2018 I was

I got the Canada stopped software developer 30 under 30, so there's like a small intro about myself

as I'm not showing my camera. That's why I just like put a photo of mine on the slide so that you

guys can get the idea that I'm not an AI or bot

I'm just a human just like you guys. OK. So I wanted to put this slide

like I'm gonna like RJ everyone to donate as much as you can

So as this discourse is really noble and we are supporting to

fight something like which is like beyond our like that. We can't even even seen where it's going and I'm also going to

thank to all the sponsors to support this this noble cause and

like making it happen in a week. OK, so now the question is why is AI important

So we are hearing a lot of hypes and like a lot of things from last few years about AI

but is it actually true? So if we look back at different waves in our IT business or IT industry

then we can see that there was the wave of software development

desktop software development, then we saw where applications or website, then we saw mobile, then IoT and AI is here today

So back in 2008, 9 or 10, when our web application, Dexter software were ruling the industry

then mobile application came. A lot of people were talking about like

okay, who is going to use mobile apps? Like who is going to use this mobile

No one is going to do that. But now that you cannot think like

your life without mobile apps, like you are using like mobile apps everywhere

like from banking transaction to see the weather, to have conversations, to do anything actually

like have communication through different social media. So that's where the AI is today

Like a lot of people maybe are saying, okay, AI is just a hive

it's not, nothing is going to happen, like making jokes at the same time

But AI is actually empowering all the industries that we have. So sometimes like some of the industry leaders

thinks that AI is going to be bigger than Internet. Why? Those mobile web or such like desktop software. They were just like a single media, but AI what is doing

It's not a single media or single platform. It is actually connecting all the platforms that we have and empowering them so that they can do beyond what they're supposed to do

So that is the reason why AI is going to be big in future

And as a business leader or someone who is involved in IT business

business. This is the right time to invest and this is the right time to get the competitive edge

So again, is the hype real like as we are saying a lot, say hearing a lot of things all on a sudden

Yes, the hype is real, but at the same time, there are some reason why we are getting all the hype all on a sudden

So if we look back at 5, 10 or 15 or 30 years back, so all the AI algorithms that we are using today

Suppose take the example of neural network or decision tree. They existed back in 1990s

So why all of a sudden like a lot of hype we are hearing

The main reason is the computational power, computational capacity, Cloud and GPUs

So we started to like, there was like a big era of computational GPUs

started from 2014-15s. We were not able to process lots of data

We had a lot of computational limitations and data handling limitations. That's why we were unable to use

those algorithms to the fullest, like of their capacity. But what happened with the rise of

all those computational cloud resources, we can leverage the entire amount of data that we have

We can process more, we can find out more patterns, we can know a lot of other things

So suppose if I just give you an example, when Facebook first lost that photo tagging option

on Facebook back in 2011-12, that time the accuracy rate was really low, 60, 64, 65%

It was giving us lots of wrong information. We had to manually tag them to our friends

But today the accuracy rate is so high that it's like almost 95, 98% and it's automatic

We don't need to like tag it manually. It's rarely making any mistake. So what is happening

So it's the same algorithm. Yes, they have updated the algorithm, used new like a fine tuning process

and like with, but it can handle more data, find out more pattern, and they have trained it

for years and years. That is the reason why the accuracy rate is really high

and it is actually used in our daily life these days. So that's why the hype is real

And it started like five to six years ago because of the growth of all the like the rise of all the cloud and computational capabilities

Now, the question could be like why the AI is important, like how big the number is, how big the market is

I market global AI market was 10.4 billion dollars

but by 2025 like it is expected that the AI market is going to be $126 billion

So in that sense, we are just scratching the surface right now

And the I like the full development and the interface is just like is ahead of us

And if we look at the like the productivity, so it is also said that AI can increase our business productivity by 40%

So all those people that are hearing this talk today, like you can ask like OK some of you are from like different backgrounds from like a health care agriculture IT or something else So how AI can increase the productivity of all kind of businesses by 40 percent

So what is happening here is like AI can increase the business productivity by 40 percent

based on the growth, the forecast, new features, bringing new capabilities like to give you the competitive edge

So how? So suppose like you have a software that is in production right now and a lot of people are using it

So what is happening with the AI and data science capabilities? You can do more processing. You can find out like different like behavior pattern of your users

You can find out which features users are using the most. And based on those, you can empower them

you can focus more you and you can change the way they are like interacting with the system so that

they feel more comfortable they I stay with you they don't move like and start to use your like

all the tools of your combinator so that is the reason and you can also like bring more like

features and you can also offer new like tools like based on the data science and AI capabilities

So that's what is giving you the competitive age in the market and it is keeping you ahead from your like combinators

So that's how it is like actually increasing business productivity. But at the same time, it's also like bringing lots of automation and those automation can like save a lot of time and resources and money too

And the last one is as a business leader, if you're thinking, is it the right time to invest on AI or not

then my answer is this one. Only 15% of enterprises are using AI as of today

but 31% are expected to add it over the next 12 months

So if you're thinking ahead, as a lot of things are happening in the COVID days

you're also maybe rethinking about your focus, your future, your current career

But I can also suggest you should also be considering having data science and AI-based knowledge

And you can see that next two, three years and one to two to three years are going to be very crucial

And you will get lots of opportunity to evolve as a leader in this domain, too

And the last one is COVID. As we all know, that COVID has impacted our life

And in general, like everyone is got impacted because of that. So is AI industry got impacted as well? Yes and no

So what is happening like from our like we did some research like the kind of like survey and like after talking to a lot of like people from our industry, we got to know that only three to five percent of our like fellow colleagues lost their job because of COVID

but because of the COVID-19 situation, AI industry is seeing more investment because you know

AI is bringing more automation and all those like visualization and all those like investment in the manufacturing industry

healthcare industry construction, because we can have less interaction these days, physical interaction or human interaction

So AI is becoming more important. And even like what we have noticed that AI healthcare spending is increasing from 463 million to 2 billion by 2023 because of COVID

Even the construction industry is like getting in like a $1.13 billion boost because of this COVID

Because more companies are bringing more automation to fight and avoid all the interaction through COVID because of COVID

So yes, AI industry got impacted a bit, but at the same time it is getting more investment

more attention because of COVID. Now the thing is, as a business leader or someone from IT industry

who is working as a software developer, web developer or some other tools

how important it is to, what are the things that you need to know

What are the most known things? So the key information about AI that you need to know are

this keywords like what is AI, like artificial intelligence, machine learning, deep learning data science

For someone from non-data science, like a non-AI background, it may look like same

Okay, all of them are like the buzzwords, are the same thing, but for someone from the IT industry

or from this domain, those words are very different from each other

So if I just give an example that will clear your vision

So AI. So what is AI? So it's just like we as a programmer, we are writing programs always

But when we are able to give the cognitive capability to our programs to take decisions

just like humans based on different experiences and data, that is called artificially intelligent

So what does it mean? Suppose cognitive capabilities are seeing, hearing and understanding through different languages

So from different computer vision applications, finding a human, understanding what we are saying

having interaction, those capabilities, when your software has those capabilities, it means it has the AI understanding

or artificially intelligent capability. So what are the AI dominant boundaries and where are the use cases of AI

So if we look back all over ourselves, like where we are today

we can see our life is completely surrounded So if I just give you some examples

if you are using a mobile like iPhone or Android, and if you have a facial recognition based lock screen app

or something like that, then you are using AI, which is a complete example of the contribution

If you are having like using Alexa or Google Home or Siri

or Cortana, then you are having conversation, which is a later language processing through those chatbots

It is a part of AI. It's really hard to find someone these days who hasn't used Google search engine or Bing search engine

And that is information retrieval AI. You have you did like shopping online shopping using Amazon or other like online like e-commerce sites

So whenever you do some shopping, if you buy like a red T-shirt, then the next moment you see, hey, you should like some suggestion regarding you should buy this black band as well, which will go with your red T-shirt

So it's like personalized shopping or information filtering or recommendation system. That is also AI

So if you have insurance, you know you're paying $10, $15 for your car insurance or health insurance

Those numbers that are coming are actually predictive ysis. Those are insurance premium, like detection or like prediction

That are also like examples of AI models. Robotics, we know we have seen them in different movies or like in our real life

the brain robot brains of taking decisions are actually AI. The utility demand, suppose, like I live in Montreal, like Quebec

So our hydro Quebec, like at the beginning of the year, they based on our last one year of electricity consumption

they tell us, hey, next one year, every month you need to pay $30 for like electricity, something like that

So this utility demand it does forecast based on the people based on the usage of last few years and all the decisions So this decision ysis is also a perfect example of AI

The fraud detection, if you are doing some kind of like unusual transaction

and then you get a message, hey, is that you from your bank who has done this unusual transaction

So that is a perfect model of fraud detection. That is an animal detection model

So that is a part of AI. And the surveillance, we see a lot of like AI models or like that with different surveillance cameras that can identify humans, that can identify there is any like how many people got in, how many people got out automatically to do some kind of like identifying all the information about the activities of the work site or like of the construction site

So that kind of surveillance is done by video processing. That is another example of AI

So if you sit back and see all the examples that I've given are you are using them almost every day and all the all of them are the perfect examples of AI and that's how our life is started by AI. So it's not just a hype. It's not just something like buzzwords. People are using it. People are you are also using it. But it's just like for your own business. You need to embrace it and like make a plan for next few years. How can you incorporate that one

Now the thing is the two other things, machine learning and deep learning

For some people, they may think they're the same, but actually they're different

Machine learning is a subdomain of artificial intelligence, and deep learning is another subdomain of machine learning

So what is happening? Suppose you have designed a model that can detect

car to identify it's a car or not. So you have collected all the images

and then in case of machine learning, this is a machine learning approach. Someone needs to a human needs to come up with different feature engineering approaches like feature extraction

It will like based on the human written code from the image, you will extract how many doors are there

How many wheels are there? Is it what is the shape of the image or the object

And based on those eight to 10 or 15 features, like it takes a classification based decision, like whether it's a car or not

But in deep learning, there is no human involvement. What is happening? You are just giving all the images and deep learning is finding out all the pattern by itself

And then it is telling, hey, based on the images that you have provided, this is the pattern it has learned automatically

And this is a car or this not a car. So the difference between the two key difference between machine learning and deep learning is machine learning needs a human in the loop

Deep learning doesn't need a human in the loop other than like helping with the data or data cleaning

and in case of machine learning, you need less data to take a decision or come up with like a good

model. But in case of deep learning, you need lots of data because you are automatically letting the

model to find out the pattern, finding out all the like features by itself

Okay, let's move forward. And then what is the data science then? If you look at the diagram

then you can see data science is actually overlapping in all with all the like domains

artificial intelligence machine learning deep learning data science is the science of learning

about the data knowing the data finding different patterns of the data so what does it mean so these

are the two example of two data frames so what is happening here in the first data frame uh based

on the transaction it's a financial transaction based data you it is telling like the data science

based model is finding like these are the similar transaction they're safe but out of all the

transaction there was one transaction on monday wednesday which looked uh like uh like not usual

unusual that's why it has like tagged that hey it's an unusual transaction maybe it's not you

maybe it's like a fraud detection model has detected that one so finding out this kind of

of like unusual like unlike similarities on like unusual cases or fraud cases this is the actual

task of like like data science but again to design deep learning machine learning models you will

have to have like data science based approaches to clean the data process the data and do all the

issues okay these were the like differences and like domains of like artificial intelligence

intelligence machine learning deep learning data science so like we got to

know about the differences and like what is happening in the market so now we

need to know about the data so as we as we all know that for data science and AI

we need a lot of data so what does it mean and as a business owner or as a

software engineer if you have data if you're handling or dealing with data

what does that mean like what is that data like it's like is representing so

So there are three categories of data. One is category one, category two and category three. All the like data frame, all the like like single values or text or single numbers on like a CSV or like that kind of like a representation are called category one data

So 60% of the models in North America that are making money that are from category one

So what are the examples? The insurance premium prediction, the like a recreational ysis predicting it's a fraud case or not

It's an unusual transaction or not. Those are like prediction, regression and like classification are representing category one data

So as someone who is planning to learn AI, so I will suggest you invest as much as you can in category one

because most of the models are in production are from that category

Then we have category two data, so it is mostly audio and conversation

So what does it mean by audio and conversation? So most of us have used Alexa or even

Cortana or even Google Home in our daily life. So 25% or even like chatbots

those are the complete, like a perfect example of category two data

So 25% of the models that are making money in production in North America are from category two

So if I give you an example, whenever we are saying to Alexa that

hey Alexa, what is the temperature in Montreal tomorrow? So what it does, it takes the audio speech and then it converts into text

and then it finds out the key phases. In this line, the key phases are temperature

like it's an action, the Montreal is a location, and tomorrow is the state

So what is the temperature of Montreal tomorrow? So it just takes the three entities from the entire centers

and based on those action state, and like the action state and the time it is trying to find out like the what is the like from its database what is the temperature tomorrow

Then it will return tomorrow's temperature in Montreal is 22 degrees Celsius

So something like that. That's how the conversation works in category two data

And the category three data we all are familiar with all the video processing, image processing or image like facial recognition

So they are image and video based processing are category 3 data So 15 of the models in North America are from category 3 One interesting thing that I want to share is the complexity and the resource allocation and the

needed, like resource needed to design those models, also increases from category 1 to category 3

So it means you need more data in category 3, you need more resources, you need more pattern

you need less data and category one and you need less pattern and like risk resources

compared to category two and three for category one so these are the category one two three

whatever ai models we are designing or we are planning to design they will be from one of this

category category one two three now if you are like a business leader or like a software developer

if you're like and you're planning to like switch your carrier or like add new expertise in your

carrier then how you're going to do so as a business leader in your software development team the

typical roles that you have are front-end developers back-end developers full-stack developers database

admin software architect and like team lead and q a like who a software tester so these are the

common rules that like the roles that you have in your company or in your team so do you think

these rules are enough to like like does all the jobs of data science and AI

let's see so this is like the next slide is representing the complete machine

learning pipeline so what does it mean complete complete machine learning for

any kind of like a machine learning or data science based model designing we

need to go through all those steps so this is where your raw data's are you do

the data acquisition once you do the data acquisition you will have to do data processing

data feature extraction scaling normalizing all the data data wrangling once the data is ready to be

used in the model then you need to create the model uh you need to use in different kind of

like mission learning algorithm uh to create the model uh that goes with your like mission like the

approach and then you will uh validate the model do hyper parameter tuning and if you are comfortable

with the accuracy with the result with the with everything with the outcome

then you are taking this model and deploying it. So do you think all the like seven like like all the like seven roles

that you see in your traditional software development team front and back

and full stack database software architect team lead or live in QA

are capable of doing this operation? Not really, so that's why like you need your

like developers needs to learn like of the game they need to learn something extra they need to

know something extra to uh like to be considered as like a data science developer or like to

contribute in this in this domain so okay so what are the roles that have evolved evolved in the

industry uh with the uh with the involvement of ai and machine learning and data science application

designing so what we have seen that these are the eight roles that uh have like that has evolved in

in the industry in last few years. So some people may think, okay, the data science

and all of them are looks like the same one, but not really

Their job roles, their capabilities, their categories, all are different. So the data scientist, what his task is

he is the one who is finding the pattern of the data, extracting meaning of the data

Data engineer, he's the one who is preparing the data, doing all the plumbing

so that data scientists can use that data. Big Data Engineer, if you have multiple streams of data

then someone needs to design the pipeline and so that your application can hold that much data

can process that much data. So that is done by the Big Data Engineer

Data Visualization is a clever. Before you design anything, before you do any kind of modeling

firstly what you need to do is you need to find out like this data has any value or not

has any pattern or not, or even like during this COVID time

We all are seeing all the graphs, all kind of like different charts about COVID and infection

So those are done by data visualization developers so that you can find our patterns

You know, like this data has value or not, what is happening. Then the machine learning scientists

All the algorithms that we are using are actually designed by machine learning scientists or machine learning researchers

researchers so they are the one who takes your data sometimes build custom models custom application

for your uh like for your business or for your need then ai and machine learning engineer this

is the best role for your uh like full stack developer or like your software engineer to be

like a or machine learning engineer what they can do they have the software engineering experience

they have the software experience knowledge they can learn about like learn different uh platform

or different like frameworks. And at the same time, they can learn how to use those algorithms

how to use, like do those training, handle that kind of like data

And then evolve as AI machine learning engineer who like takes care of the training

who like deploys that model in the production and incorporates that model

with your existing web or mobile application. So that is the best role for your team member

But at the same time, rest of the two are business intelligence engineer

So lots of BIs and like incorporation are done by business intelligence engineer

to find out different patterns of your data. And the AI architect, it's also like a very dream role

So just like a solution architect, application architect, or enterprise architect, if you have a big application

and having like adding lots of like AI models in your web or mobile, then you need to know

how your application will handle that much data handle that much load of like like a prediction and inferencing the ai architect will be the one

who will design or like overall plan or are the architecture that like through which you can add

like ai models with your existing system so these are the roles that have evolved in the

like in the industry over the last few few years data scientist data engineer big data

engineer data visualization developer machine learning scientist ai or machine learning engineer

business intelligence engineer and ai architect so they're very different from traditional rules

but uh like as market is going forward uh these are like roles will have more value and like more

jobs uh in the market as well now as we know uh that traditional software development rules are

not enough for our like ai and like data science based application designing but you can also ask

okay what is the uh like a perfect uh into end data science or machine learning uh life cycle

looks like so this is the uh complete example of like into a machine learning and data science

life cycle there are we can uh like overall like divided into three different uh categories first

one is preparing preparing the data second one is billing and training the model third one is

deploying and predicting the predicting uh based on the model okay so what is happening here

First, you have the data storage from different streams, and then you are doing the data injection

After doing the data injection on the preparing the data, As usual, like lots of data transformation, data validation, data processing, data featurization

So you could ask, what does it mean by all those data transformation? Suppose you are designing an insurance model that is predicting what will be the insurance discount someone will get

Because it's a COVID time, everyone is not driving, everyone is mostly like because of the lockdown and staying home

So why would they pay the exact amount of insurance for your auto insurance? So, suppose you have two types of data. One is like you mostly use the data. So, on the user table, you have one data of date of birth. So, that date of birth is an infinite range. So, it's like someone can like born, could born in like 2000 or even 1970s or 1950s. It's hard to group them based on just date of birth

So, but it's easier for a model to process the data and group them based on the age group

So, taking the date of birth, converting them into like finding out whether it's like a string format or what is the format of that column and then converting it into the date time format

And then like subtracting that date of birth from today's date to find out what is the age of that user is called data transformation, data processing

and the value that you are doing after all the transformation, that exact age, which is 55 or 45 or 25

is called data featurization. Okay, once you do the data featurization, you have supposed two types of data

One like two user data, one is someone who is using your insurance from last 10 years, who is 45 years old

And another person is using your insurance last one year but is 25 20 years old okay the first one who is a very experienced driver have

never incorporated any uh like uh accident but the second one who is a new driver had like three

accident in last uh like three years okay or one year now you will have to like select the right

algorithm here as you have the data you will select the like best algorithm when you're doing

the algorithm selection uh it's like a recreational pro like uh like based uh like problem where you're

predicting what will be the like optimal insurance discount for these users so

there are so many like hundreds of like recreational models in the market so you

will have to your data scientists or your like machine learning engineer will

have to find out which is going to be the best model or algorithm for this

specific problem sometime you need to like train multiple models to see impact

of different kind of like data processing of like and the model based

calculation to select the best model once you do the model training then you will have to do the

hyper parameter tuning you can also ask like what does it mean by model training and after the

training we are just getting a binary like a file what is this so what is happening based on the

model training the data that you have provided uh the model actually like we just uh crawl through

that data all the like data data frame and find out different kind of like leaves nodes trees of

like a full comprehensive like representation of that uh of the data and then we stored that

representation in a like a binary file uh it could be a bkl file or python or something like that

and whenever there is like a new data we just like go through that representation or that kind

kind of like a tree to find out what will be the value of that prediction

So that's what model training is. It just like whatever data that we providing it like converting that data into like a complete tree based representation or this kind of like a binary presentation that crawls through the data and hyper parameter tuning different algorithms have different like Lambda values different kind of like hyper parameter We need to train them like attune them Sometimes we need to come up with different combinations of hyper parameter to find out the best combination like and select the best one that is giving you the best result

OK, once you have trained the model, once you have done all the hyperparameter tuning

you need to test the model. Like how much resource it's taking

how much time it is taking to predict that one, and how good or bad the model is

Whenever you are designing any model, you need to divide the data in three parts

One is 70% is as a training data, 15% as validation data

and 15% for testing data. So you will use that testing data to test

the accuracy of your model, whether it's giving garbage values or not

Once you validate the model with the resource allocation and everything else

then you're comfortable, this model is not garbage, it's giving you good result

Good result, what does it mean? Suppose that old user of your insurance premium who is 45 years old

if he's getting 30 percent discount and the new user who had few accidents

and using your insurance premium from last two years got like 10 discount then it's valid it's

justifiable then you can say okay it's not giving garbage result it's giving like right result then

you can input in production but if your model is giving 30 discount for the new user and 10 discount

for the old user then you can tell that you can see like okay there is something wrong that there

is issue with the data and the data like the model is not properly trained so you cannot put it in

production then you are putting the model in deployment so what does it mean by deploying

we take the binary model wrap it in an api put it in a container and then host that container

docker container in a like uh uh in a like production environment and then uh we just

expose that api to our web or mobile application so that from our waiver mobile application we just

get the api call we just process that one and give the response of like the discount number

like 15 20 30. uh so this is the approach of model deployment uh now you can tell like like how many

models you can have in one mobile or wave application you can have thousands of models

suppose like one of the example is uh one of the very uh like uh like prominent uh oil company in

US they have 9000 models in production so like you can see like how important it is to automate that

internet approach otherwise it would have taken like taken them like days and years to take the

like uh do all the deployment training and deploy it uh in like uh in production environment and

incorporate that one with web and mobile applications and if you have lots of data you can

also do like batch prediction to annotate them and like do other processing and the last one in the

end-to-end data science approach is model monitoring so what does it mean by model monitoring and why

do you need to monitor the model so models are valid for certain time and after that it starts

to give like unreasonable or garbage results so suppose that insurance model that you have trained

or designed is giving the right result today but one year from now on if that um experienced driver

in like have like experience or encounter three or four accident where the new driver didn't have

any accident in that case if you if you're giving the same kind of like uh result like 30 discount

for the new driver like old driver and 10 discount for the old driver then that is unfair you need to to monitor when it giving garbage values what are the values that you getting to some different kind of like a representation to find out the pattern

of the data or the new data as well and then you can tell you need to retrain the model uh every

now and then with the new data and then it your model will tell after the retraining based on the

monitoring hey the new user is getting 15 percent discount because of his good driving of last one

year but the old user is getting 20 discount or 25 percent based on his harsh driving of last

one year so that's how important it is so this is the end-to-end approach preparing data built and

trained model the plan predict model so this predicting uh preparing the data takes 45 to 50

of the time building training model takes 25 20 to 20 25 to 30 percent of the time and this

deploying and predicting model takes 15 to 20 percent of the time in the overall approach

okay now the question is you got to know about uh the ai uh like team roles uh ai market and uh

like what is the ai application development life cycle looks like now what is the market trend is

happening in north america so there are five five times the four types of trend that we have

have noticed in North America. First one is platform as a service. So like just like Amazon

Microsoft, Google, and at the same time like different kind of like auto-mail platform

Kepler, data robot, data IQ, H2O. So what they do, so we know about the like Amazon, Microsoft

and Google and their like cloud platform. But at the same time, what they're doing

we're hosting our web application, our data over there, but they're also giving us the opportunity

or the extra facility of hosting our model, doing lots of training of our model and of our data

on that kind of platform. And they're offering it as a platform, as a service

to use Amazon or Google automated machine learning platform or something like that

So this platform as a service is the first trend that we have noticed in the market

that a lot of companies are following. The second one is AI model as a service

So again, those three companies, my Amazon, Microsoft and Google, they found out they came up with some of the three like key approaches of AI or use cases based on vision, speech, language, knowledge and search

They find out like maybe 25 to 48 like like use cases of basic use cases and then they have trained those model and then hosted them

So why AI model as service as a small business owner or someone who has like a small web and mobile application

You don't need to like do those things like just for like a facial recognition model

You need to do you don't need to do it from scratch. You can just like take the photo from your mobile app and then use AI model cognitive service as a service and then send the like your photo that you have taken to that API and that API will return

It's a human face or not. How angry it is. What are the neutral like all the like facial expression and everything so that you can easily incorporate that one and you don't need to like spend that much money as well of all the like basic like use cases

So AI model as a service is very popular as well. A lot of companies are using them

Third one is internal R&D and AI. Yeah, so hello. Yeah, yes, sorry to interrupt you

We are having only five minutes left for ending this session. Sure, no problem. Yeah, five minutes will be enough

Yeah, perfect. Thank you. So the third one is internal R&D, where like a lot of banking and like insurance companies are actually going like because they cannot load their data upload their data to a public cloud They host their own R team IT team to host like do all the like machine learning based application

designing statistical correlation, finding like and hosting their own application as well

And the fourth one and the last one, which is custom AI application, where companies like our OSDIA is like playing a vital role

We are like make like collaboration with industry partners and then we based on their custom like need, we design their approach like from scratch to find out whether it's possible or not

Finding the ROI and like giving them like that. OK, the sense of understanding that it is possible and what will be the cost and what will the maintenance and everything

So one of our application like very common example that we have recently developed like for a

Protégébu one of our client so they used to take like collect information of all the products that

we use daily they used to manually save them like this but it used to take a lot of time a lot of

effort a lot of human effort as well which was costing them a lot too but what we did one of

our custom AI application we like used a bunch of like AI models with an

incorporated them with a mobile app that can take photo and then automatically

based on those AI model as a service find out what is the name all those

like ingredients and all the manual approaches that used to take a lot of

time became like was done in few seconds through this automated AI

application so this is like a very good example of our day-to-day impact and use case of ai

that we are doing so and another thing that i want to mention before i finish like my presentation

uh what we have noticed that a lot of companies are doing just research not the like product

development so finding the return on investment is very important in this ai industry that's why

Not all kind of like cool ideas are end up in production and makes money

So you need to find out if you're thinking to design a application, you need to design a small POC proof of concept so that you test early and fail early

And you can also come up with like, OK, this application is going to cost us $10 to develop, but the return on investment is $2

So it's not like valid or like it's not like reasonable to build it

But at the same time, you need to be strictly ethical and unbiased in case of AI when you are designing it

And from the first day of developing, you need to have a production ready mindset so that the moment your model is ready, you can put it in production

So that's how we design our applications. And again, you can do lots of research, but you need to incorporate your research as a model with your mobile web or cloud applications

if you don't do that, then there is no value of that research if people cannot use it

So customer application like these are the six things that you need to keep it in your mind

But I'll do that. If I had time, I would have talked a bit more about our in-house AI product

which is Workplace AI and ID for Constructional Project Management System. So as I have just a minute left, so that's it

Thank you for attending my talk. I hope it was useful. I tried to cover a market trend

What is happening? All the best cases like how the market is evolving and where you are using AI in your day to day life

If you have any question, shoot me an email or contact, connect me and LinkedIn or Twitter

Thank you