Session Information:

Challenges in realising AI use cases for enterprises and the way forward

As we know the success rate for AI projects are very low, many use cases reach till successful PoC phase by creating ML model and desirable outcomes based on sample data; BUT moving them to production with real data starts coming from various sources ex, S3, ERP , Social media etc, and then the challenges start.

With changes in the model during training, accuracy in Inference goes down, versioning of dataset, model explain ability becomes important. This session discuss systematic approaches to take for scaling AI use cases.

Session Transcript:

Alright, ladies and gentlemen. So I am very excited to bring our first speaker today. He's coming directly from Bangalore, India, where he leads SAP's Artificial Intelligence efforts.

I'm talking about Rahul .... ... is here with us just talking about challenges in realizing AI use cases for enterprises. And the way forward rao has been the first, before he's, has deep understanding and expertise, and practice of this concepts. It's always an honor to be with, to, be with you, Raul. Raul has over 20 years of experience, and has been with sap for more than 17 years currently. He's responsible for the Artificial Intelligence platform, for, realizing, scaling, AI, and machine learning, use cases for industries and line of business. He has extensive experience in design development and product management of enterprise software products.

He's playing a key role in the Nasscom Cove with 19 industry task force to build Cove at 19 data, platforms for India, And he is also involved in government policy reveal, and publish books on AI for school students with the Government of India, ..., Go Itau, Tinkering, Lab.

There's always a pleasure to have you with us. Thank you for sharing expertise of our global audience today.

No, Yup. Thank you.

It says, my always pleasure to be part of this event and sharing my experiences from the industry to the broader audience.

Just get to hear me.

Yeah, I can hear you now. Just waiting for your screen share, and you can go on.

Yeah.

Waiting for my screen share option, OK. Yep.

It has come through now OK, excellent So you can able to see my first screen video Yes we can.

Good. Excellent. So we're going to talk about the challenges integrated adoption.

At the same time, just to take you to the sneak peek of the trends. What's going on on the EA areas for maybe 20 21 of the total equity that we started with a lot of pandemic issues but we're going to touch upon us to be on the how the enterprises with adoption. Adopting the AI and the challenges, they're facing experience.

What I bring out to be here with the role that I play for recipe in the space of AI now, and also with many customers, as well as be upwardly platform, also, working with the government agencies, and different places, as well. So, let us look at how does it evolve during the during the 23, D three D, and no priority is getting towards the tricky part D, But what we see the ... did he taught us that?

How does the AI to be, to begin to protect it and make it a little more smarter than what we anticipated?

So, before, I think big, 20 19, we always do in order for prediction, we need to have large amount of data that to be available, we need to have access to the huge amount of data sets. And it is not possible that we could able to manage things of the smaller data data sets.

I think one thing that didn't happen is that because of pandemic, the lot of areas work are disrupted, especially supply chain management. The prioritization, and we changed. And all that we have, the UN for the Health Care point of view are the way with the research with happening.

All been, you don't need to be work on the shorter scale of data. And what you have been traditionally happening, that's something that has been, has been maybe not applicable or maybe the short span of greater than the need to predict what can have, but it next week or what can happen in the next days. And that's the many algorithms innovation have bought in the side of that. How I can get the small sample of data. And, based on that, how I can now build the little logistics model, or the fact that X, each of all of aids Care division, that could happen, because of what's happening right now, because of the pandemic. So, I would say that, that's the one area, has been realized. He got disrupted. But that the area is where we see the, the checkbox or conversationally become little more smarter.

In many cases, we see the usage of like size or I mean, Google has a new board put in there the intersection of a human with the system is getting redefined now.

So going forward as an enterprise application, did anyone would like to have interaction with the system, not by typing or or or interacted with that? But by the board. With an NPS or the way we talk to the systems. And system, be wanting to distinctly to understand that how that can be taken care of and execute a command that I explain to the, to the system directly. So, the chatbots are getting more intelligent, way of interest rate, is going to be more on the conversational way, That, other than going to the ... system, and as much as particular system interaction.

But, also, the human application interaction of a human are also going to teach appliances, instruction, going to change, and when you are looking at how the helix has been able to switch off switch on the light. We look for this ..., which will tell that go ahead.

Cleanup something at some period of time. So, it is all it is not possible without the watershed would happen. How did you want to interact with the ... appliance or auditing system? Electronic order moments, also, that's going to be disrupted and lot of research happening on this side, and then they are getting more and more predictive and going forward. Another area Where we see the auto correction happening So, it's a water shed with the robotic process automation. Mimi enterprises been doing it: surprises, your enterprise is not think in this direction, but many places where there are the manual upload widths are happening. Or do you have? Well, you know, that next job, which is going on, where somebody getting, let us say, the, just to get the e-mail, saying that I want a particular laptop in a particular way.

Is that, this is a self interpret, that information, and create directly the purchase order, or rather than, there are a couple of colleagues are automating and looking to do that. So, there have been a lot of, lot of innovation happening, how the system, called the data is coming. From system getting depleted and to go to the next process. Which could be also automatically happening, even though the creating the purchase order, creating the requisition, creating the goods, because these high document that were there, that have no no border. Istio of hand-written documents, are also can be readable by the system. And they can do the activity of, the next, as statistics are recorded them. It can be coded for, for them. So this area, which reached there, we see a lot of changes being coming up now.

The another idea is that the data inability level of the, how this equal to T can be made available, because data is the important part for. And we talk about the E I N E. I cannot be produced in a good algorithm, or, or it cannot be make any end to end business processes possible, without you have good access to the data. And this is the problem which being quite a lot upon there are, a lot of algorithm gave. it pushed the data level, how the data can be, bring it together in, order to ensure that we get the right level of algorithm produce and and also the accuracy would be provided based on that.

Important part is that is Theta. Ubiquity is the important idea which cakes Quite, I would say deletions, we sub disruption. But we say, 20 21. Part of algorithm going to come up, how they can bring up the data layer, much more stronger and millennia algorithm can be pushing the data layer. So, but the model that, that, we build on top of it, what machine learning and AI, it could be very seamlessly to be used for us.

The important India, also, which is very, very close to everyone, and we have to bother about it, is the ethics, or the responsible yet, because the small human error in the board can make decisions, and maybe the million of transaction can go bad as it, when we start trusting, the, more.

Or we die, the self learning or retraining algorithms more into the models or retail system. What the right way to do it. It's also important that we have to be judicious about the issues. We had to look at, that homeless system could do it, the other less of the human interaction. Now, or are there any part, which getting back to the wildlife?

So, we had to really take a conscious decision that we had to how the AI could become law, you don't say, fixed level and that's a debate that where they could not, should I be the stepping out from the, from the ... was the the what we want from the ethics point of view? Very soon, there has been talks about that and it's a big pivot table topic that I was a antics needs to be set up. But there are, the government policies are coming out there, the companies, as with building their own policies around that as well.

And they are looking for that, How the, how the AI could could be a more potentially restrictive in the cases, or how it could be the goal at some level, or at least. But, yeah, this very important topic that it has to be used. Right. Yep. So, looking at into the, to the next section that we talked about, the trends going on, on the EITC.

Here, we talk about the enterprises, right, and most of us have been on, the enterprise is being envisioning either on the journey of AI. Or, we need to look at that. How the AI to be adopted into your organization. Maybe some high level strategy that I can show the night on. But, as you'll see, we could have more like a discussion going forward to make it ends up more details.

So, first and foremost, to understand the value of AI and how the adaptability of the Yale ... to be there.

So look, something which is just constantly changing nowadays is the, all the systems around the state, so when we talk about the real time, Real time situation of a face or we talk about the advertising business processes, and it wouldn't be channeled be changing or ecosystem being disrupted with a different part of a cloud or on prem to cloud disruption is happening on the system point of view. That is very important that when we built the AI in today's environment, tends to be thought through promise scalability point of view. Because many of the genes that happening around us is just a construct, that we are not the only to assume that is going to be, you know, they are, but at the same time, P I. To be adaptable, efficiently and innovate. You're not ever more important in today's environment.

Which is, which is the key for the enterprises to to think about, if we ought to be something, which is rigid, considering the today's constraint, is going to face challenges going forward.

So, we have to consider about the ecosystem, which is changing, and how does this, how does the solution that can be paid, or capitalize, or this approach it, from a scalability perspective, doing it?

How does that, you know, what do we re-invent the businesses that are high level broadly, from the perspective of it as the three categories, if you look at it? The first category is that we require the function, so that's where he talks about what technically the deep learning model, the machine learning, the scalability across the IT could happen. There are many open source technologies are coming up to, how should we? Should we create the more efficient modern out? Or, how does, the Monday said the model can be picked up, ... model. That case is something we called as, the statistical models can be paid, And so there are a lot of code functionality is being developed and make it available to us. So that's one area where the lot of new innovation happening.

Second area is that as we see, the business intelligent process automation and the digital area where many mundane processes are, people are looking at that, how we can use the machine learning and AI to improvise that, how can get them more profitable, oh, I couldn't get them more efficient with that creative processes because of multiple approval. It was taking 45 days and the manual intervention, people are sitting on the pile of document to make it happen.

How can just, I can take the scan of the document and the scanning can provide it to me. You can immediately create the new process automatically, rather than someone else to read the code, the data, into the system, to create a document and then how the approval can be taken. These these other areas, which are more disrupted we'd see or getting disrupted right now. And again, we talk about the how the Nixon should you whites or interface are really going to be change.

We see already the, the difference, the way we interact in our day-to-day life, going forward, Could you be, a company, has, bought two, interact with you. They want to reduce the human resources, in terms of settling the basic queries to you.

But at the same time, also, the interaction with the system is going to be generated by the baby. Interact with inhuman system, understand your language very easily and provide you a solution rather than you do a lot of activities by that. By yourself, which takes you some more time to figure out things. So, the user interaction with using the ... party to change, and that is exactly the huge amount of, the, of the market that we're talking about. And it's also the enterprise market, which is about three point nine trillion, just predicted. I will take you back, which, eventually getting to the, to the next level that will see by 2025.

There are definitely an opportunity. And especially when you look at the potential, unfortunately, probably, enterprise point of view, which is the key functional area that that, that we like to see, what is the supply chain, management is the area where already the, actually, the 50% of the large, global companies are using AI and the write once. And it takes IOT supply chain, operation will be going to be scale up, quite, quite extensively.

Second, delays, procurement. So we see the, how, the notch and the individual task involved to the source to pay their potential to be fully automated. And that's where the chalkboard or kind of, things are going to become a teacher.

The third area is the fintech that we have been talking about, and every company will still be here, and the automation chances are very high in this particular industry.

So, these are the areas where the technology and the potentially change there.

But, it's very important, also to understand that the, the ... is, it's not easy to talk to achieve. Artificial intelligence is not just to be, you know, there been a problem in hand, and just like, but you don't take it for what? It shows you a lot of challenges, and that's where we look at it. There are about, the 85% of the AI projects, speech struggled to make it to the, to the production.

There are the areas, because all the people feel that, I have a good amount of complexity in the system. But always achieving. That is very, very, very, very narrow jumps, same thing. That many organizations feel that that is the complexity that they have in order to achieve the AI because of the failure rate is high.

The data is typically built in the traditional way and that traditional data is not not bought into.

The system behave, the way I want it to conceal data is segregated into different different forms, that getting that complexity, that you have data coming from on prem data on the cloud systems, data can be popular.

Traditional system by which they would also trigger coming from your IOT device as well.

Should I blame that data to get it in order to get really important part that I want to consume?

Importantly, also, there is a shortage of resources. I would say still last 2 or 3 years, we have been the i-tunes last year.

The really, very good, where we did a lot of education, everything around AI, and we see a lot of new data science provides been coming up, there are engineers developed.

But still there, that need to be, have seen the deep understanding of, how the DEI Project success, from the bottom network, in this context, to, especially, to handle this challenge of, how should I, how I could make the project successful, the heterogeneity. Imagine our business processes.

And the imagination of our business processes is looking at first three of how can I maximize the value from the data that, that, you have.

right? So, I know my data is in the bits and pieces as a multiple places at the level, using the raw format, how I can make it more readable, data, how can I create data together? How can I connect to the all the all? logged in them together and unify them in order to someone can able to understanding how it can duplicate data are getting more deaths on the legs of the data and getting all the data together. So I, that's a strength that you have to focus on and that's a bigger challenge is that you have to build it before you think about making the EI.

Yay, pigmentation.

Once you saw your data on them, you have to look at that, How do you implement the more confident and Bridget DEA? And Machine Learning models also understand and explain the worthiness of the recent, and also automation, or the scale of it.

So, once a day, it shouldn't be only you don't make for a particular business you need on particular dealership or a particular country, then making it had to also see that this is going to scale exponentially because it's all the data centric solutions of the data centric solution has to be bulk on in conjunction with the data objects that do that, that we use, too.

So, it has to be work in conjunction with the other data sources and scale horizontally as an admitted to acquire. And also, the model has to be trustworthy that it should be extended. But, if we say that it is providing you a particular set of result, you should know why this particular set of result been provided and not the other way. They explained it.

Is that D by 2 E 2 MB build the build AI projects and also the last and the important part is compliance because compliance is somewhere where the people should have the authority of Theta who wanted to add that theta. The widening of the data and quality of beta should be should we Also go on is not available for everything. For everyone of those organizational data is important because not every data set should be accessible to the to everyone and it should be very contextual. Took a particular, personally, Arctic, but it can get a nice other filter on the, the, the model. Which need to consume data that only can be provided. So it ionization data governance, as well as the compliance data is very important part, when you're dealing with the project.

Otherwise, it's very easy to use the person that drive data of someone that can be, can be misused by by someone else as as ....

Um, when we look at the DEI, which are the personas, are typically impact data across your enterprises, so AI across it. Which is to the, to the different personas in your organization.

Importantly, the CIO who is kind of struggle with the with the requirement that that he didn't attack on that data science disguise. So critter science is always wanted to work on the floss evolving algorithms. They want to work on that new unquote, how I can make it the best open source or go to them? Can we use and good prediction much faster? We can get the better prediction in this case. You have IT Engineer, or what we call it, as .... He's the guy who always wanted to put things into production. But he has not facing things before. Because, many times, the data science guy, you typically work on the book on, based on the, whatever, the open source. Our documentation, and, and, and, and yes with him.

So, these are the challenges that the digital persona who will look forward to, but, yet, before getting into the, what can be done in order to solve these challenges?

But to understand, what is all about, is the enterprise AI, right by the top of the Enterprise Artificial Intelligence? What is Artificial intelligence that? We're talking about it? The enterprise, Normally. Intelligence. So, it's really important to understand that concept.

First, before we get to the solution of ..., or some of the approaches or strategies, that maybe, there'll be a talk about now.

The importantly what we need to understand is the the enterprise AI is you have to think through from the incubated perspective. It's not that you only build your model, to show some good results. Everything looks cool. And, then went to ..., it was in its own, right.

So, Enterprise AI is something that you build a project from, the data.

Your data is coming from, how your data is going through the pipeline, how you are getting to, voluntary ties getting free to you, or your, your models. And how does that, that model, Caden kid could expected data to your business processes. Because the end of the day that every project that you do that has to be disturbed to your business processes and business processes are important, in this case, to be to be adopted.

So, when we talk about the anomaly, or maybe to create computer use approach to get some data in Excel sheets, chosen could output, and then solve the school. But even to think about the enterprise here, it has to include from the business point of view. and the how how it am I going to manage the value of the AI, based on what business processes that are going to impact.

And the value has to be ... from their, not from the, what the outcome is going to be there, which can be also, you can check Knitwear excellence. But that, that value that you are getting, that Katie ..., from the, managing your data, Managing your, your development of, your model As well as improving the data unification and the consumption of your inference of your AI.

Do your business, versus Maybe you are showing something on the, on the Dashboard analytics, could be your consumption audio, getting some are the job, or you're making that could trigger some other business processes throughout that.

But, yeah, this has to be Intuit come together in the application. The one place that to realize that.

Also, importantly, we need to understand what is the information management process, right?

So, how do I discover and refine, Compose, and conduct an orchestrated the simplest way I can go on, and so I need to have a tools, which can work with my normal variation of a data, which is coming into the data, is coming in from the, your on prem system, your online streaming system. The social media's or theta is also coming from your cloud based systems. Now, how can I discolor, and then get the data from all those things? So I really need to have a truly, which can support in that way, how I can compose and conduct the orchestration of the data. So not only bring the data together, but also to cleanse the data there. I need to do some kind of the wood, the logic on top of the data. We are where the bike lane that and creating can be disturbed or are disjoint such a way that it can produce the right output.

And finally, I'll, I can go and compliance to link with data, that I built it, how I am going to make a government around that. So, the right data is available for for the right audience and not see that, but that's that's called infusion, pumps and management system that you can think of building it in order to make the project successful.

Um, let's look at how do you re-imagine what AI business processes, or I would say, the adoption of AI business processes.

So in this case, firstly, you need to have a foundation of the I suppose, is like you to think at the assembly line and for us to get the data. So you get that, because you get this stream data, structure, unstructured data, coming from there.

Ben, you would learn so that you'll have maybe you've used it, will do orcas data to whether big the Agile development process. You create a model that, that's what you need to be paying for that picked out in order to manage it and see the ... technique that, to your AI projected quiet. But you, you train the model. You build a model, and then deploy the model. So, that you think about the deployment of the model, you really bring the scalability of the model. Review the model performance on that basis, like the refinement. And they don't speak to the model, and also, very importantly, to poverty, How should you replaced, and editor at the model, as well?

Because, as well as you're monitoring your model, the project can be successful. It can't be successful, is only you deployed once, and just leave it as is. It's not like a non one software development code as you build an application. And that application will try to serve you for longer time, But, in this case, at least you, if your data is changing your training data, what model you have trained? And your accuracy is not getting accurate than your project could get for that.

And very important part of the connect, because this is where, where you would have a conception of your AI to your business processes. and until the time you don't connect them together with the business processes of value that, that you'd like to kick out.

This is not going to be complete, and that's where we talk about Enterprise AI project, feeling many places, where we see that either, that you don't have all the pieces, the diagram, as not thought, through completely, from end to end perspective, I did you just build a model or do you just pick the data. But when you think about the company doing business processes, the chances of success is much, much higher and that's where you can go to, to understand it better.

And, if you think from this from beginning, I'm sure you're all the personas will indicate their assets to know that. You know, you know, upfront that, how much cost that going to be.

Make it happen. And, you know, what kind of tooling you are going to use it, and which will deploy, and how this complexity to be to be sold.

At once quickly, wanted to show you an example, from the real-time manufacturing, it just gets its ease beam into into the mode of business processes area. This is the example of the predictive quality manufacturing plant where it's typically the, the from the molded product comes out. The product has to be, or what is the component that comes out. That has to be tested automatically. Definitely, because manual testing of the tipi strategically in your car. With that said thousands of this, we get created every day and taking take the quantities are deficient. In this case, we did really looking forward for getting the IR images, these typically tell me, the component has been viewed as per the dimension or not, and there is a machine learning model, which we run. to compare that with the, with the visual.

Get validation, that the all the, the model has been created as what they expected, image. Then we have our pressure data because we checked the weather.

These particular and then put a pressure sensor pre chevron on this particular component automatically. Stating that the material use it is been used so, it can use in the correction. But it used to be and then also to prevent this data with my material master.

What happens if, I don't think, from my end to end perspective? I can create all these components in silos because they use different technologies.

At the moment, I can create these sensor data coming storing into the ... regular bases, desire to be a media company because they are able to be able to store stored in the cloud. And some are Modular unit, which will give me the ... comparison and give me the output update.

But, yeah, again, due to the prediction based on the biomass theta that I have coming from ERP, I need to have another system to get connected safety. All the system, what this joint is not that I'm going to get a complete overview of them.

So, what I need to have that I don't think from a big thing that how this whole system, then how I am going to go to create these data pipelining. And then the inference of the activities and the predictive dashboard. Aquatic yeah. That are going to make it. So I would get these Kafka. Went swimming also to the cloud framework by the data pipelining, The data that's coming from my eye or images, the inference of data, the data, which is happening there. And the outcome of this both takes I was saying to the, let us say this case, the ACP hana system, which basically, labeling these data will be even looking at not only the current quality of Dakota, but also giving based on the quality, the prediction. A prediction, great dashboard for my usage. So, this is what? How would you solve the problem?

In nutshell, what? How does the project has to be? Do we bring up to success? You have to always embraced the open source technology to can't just think of, you'll be the some home-grown part, because he's been very disruptive environment. So, the architecture that, Q, You look forward for your organizations, You should always look at that something which can embrace of scale with the open source ..., every six months, It's been changing quite heavily. And E two is there to be, again, the wall document, to leverage the open source. So what today you are building, as an assumption that you are making, it may not be ready for tomorrow. At the same time, the runtime that you use, also, it should be adopt to the maybe our Python, and there are many more companies.

Very important that the layer on which you build your scalability is very important, that you should always look forward purple belt upon it as kind of an environment, which is a scalable horizontally as either as his demands. Because today, let's say you're building applications for the one system, one environmentalism company, one country, but eventually cause as to the scale out to the multiple companies, multiple organizations. And you do need to be in a situation where your application is not scaling because of the way you wrote your environment. It shouldn't be always be the kind of containerized system using Docker or Kubernetes clusters. So you can use the choice of hyperscale data but at least it does something which is the categorize, this terribly scan, scaling as and when required. So that this is the way forward for making it more successful and scalable environment to bring forward to our enterprises.

Um, definitely, how should we get started?

So you should always start, you should look at that, how you should build the, the first two are information management system. Followed by ..., books, all your data that you want to curate in order to achieve your business goals is all visible.

And then top of that, you do the initial learning experimentation, build the good keywords, The machine learning models, use cases on top of it, and then we'll build the right projects and select the right technology that support you, or the building, the AI, the output, and then you deploy this model and didn't do the machine learning on top.

We can try to scale it up at the end of the day. You've tried to augmentation that into your business processes. I did automation, audio business process to be being present. So, this is the way forward for, for being a successful of adoption of the EEA.

Now, for your organizations, and the Yeah. That's pretty much that I have to share your table when more than happy to take the questions.

Excellent coverage, Raul, I really appreciate. the fantastic. We're going to be relaying some of the questions. If you, if your audience is still has additional questions keep providing them, I'm keeping tabs on them. I'm gonna pass it on to Raul in the time that we have allocated here.

First of all, thank you for for a great comprehensive veal of AI today, and really how it's how you see being deployed across different enterprises and the challenges that, uh, that you face and the, and also the, the path forward you create. You would provide a very nice, best practice approach, if you can, call it that way, to how, how we should deploy AI effectively in our organizations.

What do you see in the marketplace today?

That is, one of the common themes on the questions here is, what do you see in the marketplace today related to AI?

That's, uh, that may be it's advice that's being dispensed by certain experts. But it's really not that helpful. It's maybe someone who wrote a book about AI, but someone who's not really practicing AI from a practitioner standpoint.

What what, what type of advice is either not helpful, or maybe we can flip that and say, what is most helpful for our practitioner to that that's implementing AI in their organization right now to get right?

Yeah. So as I also sharing it with my presentation as a practitioner.

I've been reading books and and a lot of TV available on the AI is is is there. I don't suggest cocoa to every them up they are. Making your hands dirty is the right way. And as we look at that, it has to show you the value of your business process, of what you should get convinced about. This particular implementation is going to give me the value for the business processes. Then you should look at that and come backward from there, right. So let's say I'm talking about the, making the RPA kind of investment there, argue you have, you know, reduce the, reduce the footprint by implementing a simple project of the visual inspection kind of stuff there.

And you look at what kind of data you have, your video data data to make it to curate your step by step and then make a bigger simple project and then see how the league the value to you. So theta is good which is available in many place. I'd take you to get started. Just just get started. The process that that, that retail suggested mostly on, the 1 by 1 with a data visualization. Make your model deploy. Try it out. You will learn a lot about your own organization and the challenges that, that, that, that your face and there are plenty of solutions.

They are always remember that you have to be use open-source, don't get to the monolithic system.

You have to think of scaling of the system. So use these hyper scholars for the co-ordinators and aqua technologies and B will be open to adopt the D word open source technology which help us to, to make it happen. Don't try to make it everything by yourself. Or everything available. Leverage them. What's the market? It will be the easy path to success. Rather than building something that would just get started reading and doing so instead he's not going to him for long.

That's fantastic. And I was smiling because as you're speaking about it, you're addressing another question that, another theme that has emerged here on the questions, which is related to proprietary versus open-source AI, and maybe what path to take? And you have, kind of indicated that already in your previous answer, And so the follow up on that is that, for people who are getting started, and that they don't even know where to start with open-source AI, are there any guidance that, is, there any guidance you can give to them about where to look for open-source AI, where they could get started?

Yeah, I think depending on your industry, Or, you know, depending on the what use case you want to address, right, there are many, many things which are available where you can just get started with. So, I would say, you how. The open source libraries come up with the also noted to leverage that are available, where you can just look right ..., are there, which, which also provide a large set of function. Depending on if the statistical problem that you are addressing, then you how you apply by kind of areas also, which gives you a different kind of laboratories to, to get started. So there are the baby projects, which are also people are contributing to the GitHub. I would say look at those, those GitHub projects, if you just start with googling around it. There are, as I don't want to, bias with another one or the other as such there. But he offline, if you want to connect with me, and I can suggest also that we do that, basically, so on.

What exactly, challenge you're looking for to guide you, to the right open source, to do, based on my experience, is there. But there are a wide variety of things available and good part of, of, of this community for an AI. Is that all built? getting built based on how people are contributing to it? So there is a solution available for everything that you taught in the initial phase. Someone is contributing to that. And then based on a problem, but there are, you just start Googling. I would say you would get A get the best answer.

But I'm also open for for supporting you, here in case specific problem.

Fantastic arrival, and the, as a bit of a follow up on the governance discussion that you had during your presentation, we understand intuitively the importance of governance for deploying this type of tools and techniques effectively for value creation.

If you can go a little bit deeper on, on the organizations that are doing it Right, and they're getting results from AI, should they, What does typically, the governance look like for organizations that are doing well?

So the governance look like at the political level, so at least they said that that I have been working with and also I can talk about our organization point of View first, importantly that the Data Access layer.

So once you make your data, the data layer ready, it has to be access controlled via the lad, the some profile, is what, what we call about. So not that, one.

Or to have access to the all kinds of data, it has to be approved by authority and also need to be required on a need basis. So yeah. If you need for some period of time, you get this thing and that you can revoke. So you have to put very strong data. Governance policies. Basically, that we make available for a second part is the data.

An organization are data sets are part of labeling and the normalization data. Because, if you see, typically, problem that you are dealing with, you need to know the last quarter of results, and particular region of reason. You don't need to know that. Who got what numbers? Or what people are doing the arc. Which store? Which person? who has bought what? We do need to get what I need to noise. Or the number of particular product. Which, hold the sale happening on what's happening at that, is for sale and that particular region, and I can profile the certain, for users based on that. Right. So, determination is the important part here also, when we talk about data governments, because I could mark the data, which, which only I use for my purposes, right?

And then, that agency for, for getting the right amount of models, because then I can, I don't need to be, Be worried about that inmate. Getting into somebody, personal entity, or I just focused on this problem, and getting what data that they're that, that I need to have that, again, we create the talk about data profiles as well.

Where the some kind of, well, in the past physical Data mart, but in the modern ... type of options available, where I can load data coming from multiple, multiple data sources. But I only build, then, in a contextual way, and that also we go to the, who can see what data. And what kind of data inferences can be can be up? And never deliver data to the to the radically the end? User? Is always available in the form of the genetic influences? Or do we have in the categorization or sample of data, or stuff like that.

So there are multiple ways of I've got the data in the case of AI and always data that we need to be picky. But just summarize it. I would say these are creating the access level and the data our organization is the very important policies that we should look at to get started.

The data going, let's say.

I didn't have my microphone was off momentarily.

Always a pleasure and a gift to have you share your expertise, a true leader and practitioner of artificial intelligence. I know it's, I'm broadcasting from San Antonio, Texas and it's morning here and I know it's evening time in Bangalore, India. And I'm very, very grateful for you to take the time to share your expertise of our global audience today.

Yep! it's my pleasure to be there and I could see your motion on your platform. Thanks a lot.

Yeah.

Thank you.

Ladies and gentlemen, that was row loaded the director of for Artificial Intelligence, SCP, directly from Bangalore, India to us and sharing the best approaches for AI implementation infrastructure build-up in the, in an organization like saint B, which is doing this across all their client base and around the world. So, we're going to be shifting gears at the top of the hour. We're going to be moving from Bangalore, India.

And we're going to be moving to another continent and the Italy where we'll have the leader for advanced services business development, and your knife, unified collaboration from Cisco. Josep a burrito is going to be with us.

He's going to be talking about the migration of calling and contact centers to the Cloud very heavy application of IT infrastructure and Cloud Strategist. Giuseppe has over 30 years of leadership experience in the telecommunications industry. And he will share with us the journey at Cisco that they have been through on this migration on to advanced platforms and IT infrastructure and Cloud. So, I will be closing the session for now, and I see every one of you back at the top of the hour. Thank you.