Good morning, good afternoon and good evening to all the wonderful people here and thank you for joining me. Uh I am Narita Sheil Khan.And I'm going to talk about, you know, the advanced Analytics and A I MS services um especially from Microsoft Asure and how all of these services could be stitched together to bring up an advanced and modern analytics architecture. So um let's begin. So um you know, if you could read the sentence in God, we trust but all others should bring in data, right? I mean, this is how uh we would be able to trust our enterprises unless we have the data to prove the insight and accordingly take business, business decisions. Um We would not be able to grow enough, right? And uh hence, I like this statement a lot that, you know, we we we can trust God, but we cannot trust anyone else unless we have enough data points, right? So data today is actually considered as a strategic asset because it brings in one of the kind of data when we talk about the house, it's about the whole ecosystem of the data and you consider the whole life cycle of the data.

When you have it end to end considered, you know, the data can be a strategic asset and it would actually bring out a lot more value to an enterprise. So in today's um world data is all about volume, you have huge, huge volumes of data. You you it is predicted that in 20 by 2025 we would have at least more than 175 gaby of data all around the world. OK? There is a lot of variety of data. I mean the type of data you have PDF, you have, you have video files, you have J you have JPEG files, you have, you know parking, you have raw data, you have the traditional RDV MS tables, right? And the velocity, the velocity is basically is referring to how the data is coming. Like you have real time information, you have the batch information, you have the transactions happening day in and day out and the speed of the data is actually is what referred to the velocity.

So unless you know with such a plethora of data uh and variety of data, unless you have a you know, process in place which would take in uh and have the right process like like people um you would not be able to answer the questions which are written on your right, you know, on the right hand side of the slide which talks about that.

Do I have you know, really is going to make business decisions, right? And um can do, is it really with the correct people? Do we have to correct authorization and authentication in place for the data? Do we have, have uh you know, in it fast enough to have, take you take the business decisions real time and do we really comply to the different regulatory, which are defined by the government in different countries? Right. So, so unless you have uh the end to end process to handle your data and your enterprise, you would not be able to answer these queries. OK. So um when we talk about the data which you can look at in the middle, right, the data and A I systems, right? Uh it's actually going to be a digital fact loop wherein you have uh your data and you have this particular feedback loop, your enterprise organization to basically, you know, fetch the required information as and when required. So we engage the customers with the data which you have, you empower your employees with the different insights which you can get from your business. There could actually transform the product and optimize your operations based upon the day in and day out data which you're having.

So this digital feedback loop is actually going to enable our enterprises um to to get better into their business. OK. So data driven transformations are actually giving the significant benefits and this is from the Harvard Business Review that if you have an enterprise which has the data driven transformation, it would give you 4% increase in revenue performance. It is actually uh you know, gonna give you 44% for market. So if you compare two organizations, if one is actually data driven and takes in takes decisions based upon the data, it is definitely going to have 44% higher time, uh faster time to the market and better customer satisfaction and definitely a lot of improved profit results, which is kind of mentioned as 54% increase in the overall profit, which we expect.

Now coming to our main topic like this was the background and you know, uh asure data basically has a number of things. When you talk about data landscape, it has you know, data modernization on premises today, you have the data modernization to a cloud. So here we are more talking, you know, referring to a. So you have so majorly um these two are the traditional things which are there in place. But cloud scale analytics is the modernized thing and is the environment of enterprises to basically take insights out from their part, actually could be able to enable the A IML capabilities to have advanced analytics on top of your data. OK. So we will take focus about on two points. One is the cloud scans, the part is the that based upon uh services which we have within asure and the different architectures. Well, the Azure Analytic Services, which you can see like you have the three boxes out here, the left hand side is really talking about the injection process. So whenever we talk about analytics, it's it's like first you get the data at one place, then you process massage, you prepare the data and ultimately, you're gonna utilize this data.

So the the first portion which it talks about the first box, it is about uh the injection process. So in injection, you can actually ingest from the different IUT devices, from the different events like the real which we have and you can use the data bricks or you know, you have the A these are things could be able to help you land your data and your system. OK. At one particular centralized place and from there, you would actually be able to do, you know your data operation um create and warehouse using the synapse a analytics, which is the build box which you can see use data bricks to create notebooks and have your A IML workloads, basically different data model.

Um you know machine learning models to have predictions, forecasting models, et cetera. For as per your business requirements, you can use ht insight also to take care of your big data. Then the Azure machine learning, which is, which is nothing but it would give you a capability of the, you know, creating different machine learning models and deploying them. I would talk more about it as we go further Azure stream analytics, it's majorly for the analytics of the real time information. Like you have data for manufacturing units, you you you can write it real time on to analytics on that particular information real time. That's where you know Azure stream analytics come into place. And on the last block which you see like the A R analysis of and power Bier is actually consumption wherein the higher management or the decision makers of the enterprise would be looking at this data real time and then taking key decisions to drive the business. So what are the common use cases? Um of the big, big data and advanced analytics? One is the traditional data, which is, which has been there for quite some time now uh wherein we had been used for traditional different um tools as well.

But as we go, as you go further and as, as the data is growing, as we have the requirement to be more data driven, you know, the advanced analytics comes into place wherein it is nothing but it predicts, you know, for example, it predicts the best offer for your customers um and will help you to give the forecast forecasting um of your uh inventory or it would be customer churn information which you need.

So wherein you need through machine learning models to be already deployed, so that you could take out that advanced, you know, analytics and insight and insights and incisions. The real time analytics um is basically to drive the insight from the data streams which are in real time like you know, the IOT or some information from an in eer website. OK. So just to dive in a little bit of, you know, on to each of these boxes, right? Uh modern data warehouse, like the services which we use like incision and the preparation layer, your it is clearly onto. But what you see on the left hand side is your source information that like the structured data, the logs, media files, your JPEG or you have CP files, business or um custom application data, which might be more structured data, which is coming from some database, et cetera, right?

Or usually, you know, prepare ingest using see, for example, in this case, it's the adjourned data factory which is nothing but an uh integration tool, extract, load and transform, basically extract, transform and load tool uh which will help you to do injection as well as do your ETL workloads.

You can use the data bricks as well um which is actually 10 faster than the Villa Park, you know implementation. And uh data bricks is um um you know, it it's the assure and uh you could actually leverage the data breaks directly. You get the support for both data breaks uh the your uh capabilities. OK. Then you have the model lenses X block wherein you know the synapses comes into you create your warehouse and you load the warehouse and from there you actually use and the cash. If your book, you can see the link, which actually um but all the information the data ingested from the left hand side, which you can see your sources, you actually load all of the machine. And from there, you can uh accordingly utilize the data which is required as per your reporting, you know, requirements into your synapse or into your data warehouse. OK. The advanced analytics uh in the same data warehousing ar you actually bring in the machine learning capabilities using data bricks to train and serve your models basically. So you would actually have this data which would kind of be more like in batch process. And then you would actually create different models of predictions, you know, whether it's going to be a customer churn to ba basically look at the data and find out advanced information from your within your data.

You know, that's where this uh as your data capabilities comes into picture day time analytic, which is nothing but actually streaming information. So as you even talk on the, you would see that one almost has been added, which is sensor or IOT data, right? Burn, you would get the information which is streamed and which lands into your um assure data lake um storage here using the even tub. And then this information is also fed into your um uh uh into your warehouse, you have data to process that information and ultimately, you would actually use, you know, power that well. So you can actually maximize the Rroro report of 2019. But you know, this is something which has come out uh from for the micros Azure Analytic services that you could actually look at 271% average ro I if you kind of ha you utilize the Azure Analytic services. OK. Now, um the traditional warehouse have already been there and how we could be getting better and simpler with this particular uh whole enterprise data analytics requirement, right? You you can actually simplify this analytics. You can you know use the existing skills for all the analytics and also apply your A I wherever your data is there, right? So that's where the capabilities of your synapse analytics comes into picture. OK.

So synapse analytics is a stop you know solution for a number of things for the enterprise data requirements. So basically a limit is a limitless analytic service with unmatched time to insight that delivers the insight from all your data across data warehouse and big data analytic systems with blazing speed. So the speed is the keyword here. And um it has actually a sequel and not just in SQL data warehouse, but as you look at it, um a number of book capabilities like you see the Snap studio studio, it would actually give you a unified experience, you have the integration layer wherein you can create the pipelines for the injection from your left hand side sources.

You have to, you can do the management monitoring as well as you know, cater to your security requirements. You have the different analytics engines on into Synapse. One is just on demand sequel, you know, um sequel and also the SQL provision, which is your data warehouse, which is called a dedicated pool and also the spark, a practice spark, you know, so you could actually bring up a spark machine and run workloads. Plus you get, you can also call the different workloads which are there onto your Azure machine learning and bring all of them together. Uh Additionally, which, which this particular is not showing is the data, governments governance requirement, which we have, right? So we have something called as Azure Purview which takes care of different governance requirements of the enterprise. So it is uh it is also added to the Synapse studio where you could actually create different data policies, you can have different um you know, a uh authorization requirements.

Um uh And all of that could be governed through a single portal. Uh And Purview is one of the solutions for that. So as you can see synapse analytics, you know, kind of brings up the complete ecosystem. You have the power bi I integrate seamlessly with all the assure services and you can call up your assure machine learning uh models, you can, you know, uh send information and you can create um scalable reports, basically. Um just look at if you have any questions. So moving to care is basically, um now we spoke majorly about the um architecture on uh you know, and the components of analytics, right? And now I would take you through the different uh your A I capabilities. So it actually has uh like we uh yeah, we have a number of solutions um and number of services for machine learning, you have A I apps and agents and for knowledge mining, right. So what it really means like for a short machine learning, it is an capability or an managed service which would give you um capabilities of, you know, um creating different machine learning models collaborating within your data science team or your data analyst team. And it is for all skill levels. So it gives you the capability of, you know, bringing up your compute and running your machine learning models. It might be on any of the different libraries which which you, which your models might have been built upon.

It is the industry leading ML lab. So the machine learning operations could be easily achieved by, you know, the whole life cycle of your machine learning models can be easily achieved, you can deploy them using the machine learning, it's open and interoperable and it is trusted.

So 1 million experience per month are kind of run into uh on um on the machine learning. Um basically, and this is kind of an data data, but definitely it's much more than as we're talking about it. So what are the common use cases of machine learning? Um So couple of them which I mentioned here, right, the predictive maintenance um or the fraud detection, inventory management, demand forecasting or sales forecasting, right? So basically, you would want to understand what is the demand of a parti particular apparel. For example, you would want to forecast the demand and accordingly, bring up your inventory, right? Or you would want to detect uh different frauds which are happening in your bank banking transactions, for example. So how would you be doing that? You would have your whole uh data injection pipeline. And on top of that, once you have that particular data, you will run couple of machine learning models like for fraud detection or for predict predictive maintenance. Like you would have prediction models to look at it to look at the different different manufacturing units where you are doing maintenance, if you would want to know that a particular machinery is having some issues or it's a couple of its parameters are not green and then you need a maintenance immediately before it kind of breaks down, right?

Or you need sales forecasting or intelligent recommendations. Like what are the different they might be uh you know, for one of the UK site for your Ecommerce website, uh you would want the recommendation based upon, you know, what user like if I log in and I'm searching for a red color, um, pretty dress then based upon my preferences, uh, what I have been searching so far, it would give me some recommendations plus based upon my history of purchases, it would actually give me some recommendations.

So, coming to the predictive maintenance, like you can see a, see a, uh, you know, left hand side below corner, there is a device at a well site. OK. And um OK. And uh just a second. OK. So what we're doing here is you have IUTH basically picking up the information from this particular website real time and it is this, this information, we are actually streaming real time and processing it. We have the data leak out here where we're saving the information and then you have an application or a dashboard which, which you know, would throw up the alerts or which will show the dashboards where there are some faulty issues or you know, some parameters are not in place for this particular website.

So this is kind of the predictive maintenance, you know, this is kind of the injection which happens right now, this data comes in and you save it into data stream to the stream processing and um you would show it up in your dashboards. But what happens is when you are actually going to find out predict, you need to build up this particular prediction model. So you would actually bring in the Azure machine learning into this place. You would create your own model train the model based upon what data you know, it's gonna come, you will package it and deploy it this particular model. And once this is deployed, you would actually start monitoring it. And then, and then you would actually see once this particular model has been deployed, it will go through the machine learning model, it will do the prediction. Um And you, you will get the real time information about what all you know, data your well scientist giving. And this is one of the use cases. Similarly, you look at this particular use case, it talks about the inventory management like this is a pure uh you know, pure system, any past system. Basically when you, you somebody is doing a billing, for example, in um uh at a mall uh in a particular store. And then this particular data comes into the data lake, you have the synapse analytics wherein you put all the data, the transaction data.

And then ultimately, you have the inventory ordering system basis, you know, OK, this particular appel is now out of stock. So now I need to basically order more from from a specific vendor, right? So you would actually traditionally do this. But when you want it to be more real time and do and management faster, you would actually create um machine learning model, you would do the batch influencing basically out here, you build the model train the model package and deploy it and then start monitoring it well. So this is uh these are, these are two case use cases of your machine learning. Now, when it comes to apps and you know A A I apps and agents, there are a number of services which comes under the cognitive services. So um the cognitive services have a number of different um services, different A pas which are already ready to use um which I'll talk about in a minute and you have a sure what service. So bot is especially um the service which you generally require for your customer support requirements wherein uh you would want to have specific bots to uh to answer questions to your customers or then also transfer the calls to the actual human agent, right? Uh uh to take care of the um customer uh questions. So when we talk about the cognitive services, we have a couple of areas. One is language, one is vision, you have the search, you have speech, you have decisions.

So you need to do a lot of uh text analytics. Uh to cater to these, we have to services. When you talk about vision, you have, you know, custom vision or computer vision, you have the video indexer or you have the form recognizer to read, read the data from different forms or maybe PDF documents, you have the face recognition when it comes to speech, like if you want to translate speech to text and text to speech and from from there, you would want to lose translation.

So maybe somebody is talking in a specific language and you would want to convert it in it into English, then the translation would come into translator would come into a picture. Uh You would want to do sentiment analysis, right? So you would do use the text analytics, sentiment analytics. You have web search, which is nothing but you know, you have bing searches available with a Sure. But along with that, you have cognitive search search available, which basically uh we'll talk about it in the knowledge mining plus you have anomaly detector. Um you have the content moderator to moderate the content and personalize personal uh basically to do personalization um uh cater to the personalization use cases. OK. So common scenarios are like the robotics process automation personalization, which I just spoke about object object detection.

You want to detect a specific object, say you would want to detect in your office. If somebody is coming wearing a mask or without a mask, you would want to detect the real time and then basically stop that person. So you'll use say uh you know, uh computer vision to detect and accordingly take a decision. You want to do sentiment analysis on to say Twitter feeds or um or onto the comments or those reviews which you're getting for an E onto an ecommerce website or you want an intelligent agent to basically take requirements of your customers or you, if there are any uh issues with any product and your customer is escalating.

So you basically for the customer satisfaction, you take them to a very uh cool customer support so that using your bot and then transferring it to escalation modes, right? These are different scenarios. So to talk about robot robotics automation, right? So you have different client reports, ok?

The surgery function is nothing but an cus capability wherein you have some code which basically runs. And this form recognizer which you can see it is one of the cognitive services wherein you could basically um basically be able to read what is there in this particular document which you have passed, you can actually fetch the key value, key value pairs, store it and then accordingly do analytics.

So basically it is very difficult to read data from say um um PDF files or say from from a couple of forms, say for form from a bank and fetch the information and store that information, right? So all these things could be achieved using the form recognizer intelligent customer support, which I just spoke about like you have customer call, you have somebody uh a customer who's calling, who's not satisfied, he's calling the specific call center. And then there are different services which would underneath run and help this particular customer go through this call. So there would be speech text. So it would detect the speech, uh it would detect the language, it would convert it, it would do the specific text analytics like the sentiment analysis, the key phrase traction. What exactly this particular customer was talking about? The name entity recognition. If there are specific named entities, like you know specific organization, a specific location the customer is talking about and then this this particular analysis in real time, you can actually do you know the call intervention, you can go to an or make an escalation or basically look at it in a after having this whole data data in place, you can actually have batch processing and basically look at how much you know is the satisfaction rate or how your customer calls are basically going on.

Are they really satisfied with the answers how you're supporting? So this this is how it would help the intelligent customer support use case. Now coming to the last one, which is the knowledge uh mining, right? So if you talk about the application search the earlier, it was basically more like you would search basis your keywords, right? Eiffel Tower, for example, in this. So you would search and you will get an Eiffel Tower. But now you could if if you could search for famous tower in Paris, then you would also get to know it's an Eiffel Tower, you it will automatically under and and give you uh the Eiffel Tower outputs that was a contextual search and now it is the cognitive search. So if you search for Eiffel Tower, you could actually get the images, you could get the videos where Eiffel Tower is referred to. You could get the audios if there is any music related, if there is a song which talks about Eiffel Tower or the different documents, if it's in word PDF or you know, text document or on PPT where an Eiffel Tower related information is there all of this information could be fetched if you have that in your database.

So that's what you know, the cognitive search would actually be able to help you. It is a fully managed service. Um It isn't built in a is uh capabilities which it gives you so that you could search different uh you know, data types, like your videos, your audios, your text files, your images, you could search that particular information from any of the type of data and it is customizable.

OK. So uh so as you can see what exactly this cognitive search would do, you like if you want to have the knowledge mining done on top of your plethora of information, for example. So you could see that you ingest your data from the left hand side, your data source, you have different documents. Uh you may have different data databases, your tables, all of that. And then you have something called a NRI which is nothing but A I enabled capabilities where it would do the language detection, face detection, you know, organization, entity extraction or you know named entity extraction, basically organization, personal location, key phrases in extraction.

Or if you have any AM L models already built in, you could be able to run those models on top of this particular data. If there is any translation requirement, like you would want to translate it, say from French to English or vice versa, or from Japanese to Chinese, you could be able to do that. Um sorry, Japanese to English or Chinese to English. And then you would actually, you know, create this particular map for your database and do a knowledge mining so that whenever you search, it would be able to find out um um information. And uh in no time, basically, once you have this particular indexing done. OK. So you could actually also search what is the background on the Eiffel Tower and you would uh get that information as well. OK. So what are the common mining scenarios like it is used for the auditing and compliance, whether you would want to search your audit reports or compliance, whether it's complying or not, if you have different contract information and you need to look at them that you know, whether it is complying to all the requirements with all the field or all the specific fields have been put up correctly on to that contract document, that's where this particular thing into comes into picture, you have the Digital Asset Management process management catalog search, you know, technical content review, if you want to do the review and customer feedback analy analytics basically.

So uh when you talk about the auditing and compliance, um so you have the different data set which is on your left hand side, your source, which you ingest, then you have the cognitive skills. Basically, you would extract the key phrases from your specific documents to look at the auditing and compliance requirements, right? You know specific words that you want to find out you would want to do in translation, using the translation service, you will use the language detector basically uh to uh and these are all, you know out of the box A PS which are there uh with assure cognitive services. And then you could also create custom skills. These would be kind of the customization done, especially based upon what data you have. And then once you have enrichment done, you would be able to search it um from anywhere. I mean, if you have an application from where you have created the search capability or uh you could you could have it from from your any application basically, right? Or uh for example, here it's it's talking about in web app. Now, digital asset management, you could, as you can see you have different magazines or images wherein you would want to manage your assets and have a link between them.

So here you have like the the geo point extraction where exactly the location, information, you know, biography and richer you have the image tagging like each of the images, what what you know these images are talking about. So you would actually tag those images. So whether whether it's a picture of a cat and cat is sitting on a tree, then diff different tags would be attached to that particular image. Um And there is a milk kept. So all of these things would be basically tagged, custom object detector. You would actually want to detect the specific object in these images, like if it is related to Maxine and if you want to detect say some couple of brands, um then you would be able to do that, you have similar image tagging. And so that once you tag similar images, you, once you search for a specific image word, then it would actually give you all the similar images when you're searching it, right? And then you basically use and web app to explore this particular information, right? So these are two to use cases of knowledge mining, I would just go through uh the architecture basically. All right. So as uh you guys mentioned that I weren't, I wasn't really audible earlier. Um So I would just go through this particular architecture for your reference so that whatever I spoke earlier, you could just get just I know we would uh we had only you have only a few minutes to close.

So let me just talk about this architecture which you can see this is an end to end modern um analytics architecture. So on the left hand side, you have all of the information, right? You know the big data, data, big data streams like the IOT devices, you would have the click string information. Um You would uh pick up the data from different IOT devices. You have unstructured information like images, videos, audios, pretext and um you have the semi structured information like the CS vs LOGS JN XML, right? And the traditional structured information that which is nothing but the database information correct. These all things could be ingested using different uh capabilities and uh different tools like you use in synapse analytics pipeline to basically ingest all of this information, create a data lake which you can see. This is now a batch process which I'm referring to. This is the data lake where you would actually store all the source of information at one particular place and create um a centralized repository for your complete information from there. You would actually create process that particular data and create your data warehouse, which is nothing but this particular area which you see the dedicated SQL pool wherein you would be able to fetch the report. So which you can see on the right hand side.

And you know your traditional business intelligence reports are the uh you know, for your higher management or decision makers, these reports would be available, right. Along with this, you would actually be able to integrate the big data scenarios with the traditional data warehouse using the spark capabilities.

Or you could be able to actually fetch the information and query the information from your data leg rather than completely saving and transforming that data and putting it into a new warehouse using your Soulless synapse pool, right? So this s synapse would be able to give you the whole uh uh you know, ines and transformation and loading capabilities and creating a warehouse. Now, if you look at the top um top layer where it talks about the ingestion pipe, uh ingestion of the real time information where you would use the IOT hub or even tub sorry IUT hub or even hub, which is giving you the real time streaming information, which is nothing but your Lamar architecture.

You would be actually storing that data into your data lake for your further analysis, maybe the batch process on or you could actually pass it through the Azure Stream Analytics, which is a real time Analytics service which will do your analysis and then uh give you the output into your power BI I reports or dashboards to, you know, to, to check whether there is any anomaly in your uh on any of your IUT devices or if there is any dissatisfaction of customers, this is the comments which are coming in or if there is any Twitter feed, which is actually ha uh creating an issue and you uh that needs to be addressed.

Uh It's uh so all of that could be done using the adjust team analytics. Along with that you would actually have this assures cognitive services and assure machine learning model. When the, the already trained and deployed model would run through the data, which you're ingesting day in and day out. Plus you would have the different um cognitive services to enrich it. If it is, you know, you would want to do um say further use in form recognizer to fetch the data from, from the unstructured data. Maybe the PDF S or you know, uh you would want to fetch information from images or do ob object detection. All of those things could be basically done using your cognitive services. So all all of this data, when you have the data, you run your models, you have the um insights ready, you would be actually utilizing it using your power BI I reports. So these would be your consumption layer wherein the business users would do the analytics. They would take the, you know, uh decisions real time, you would have different applications which would take automatic actions based upon what data is this coming. So that's why we call them as the data driven applications. Where in you know, if a particular threshold is going up then you need to do something on an application, um you know, a particular um uh inventory is going down.

So you, you don't want to show up that inventory to your customers, right? So different decisions would be taken up and um or you have a recommendation in Chin wherein uh you have an application which would actually real time fetch uh the recommendations for a specific um uh user of the Ecommerce website, right? And you could also use the data share.

Basically, it is to share the data set with different um with different people. OK, within the organization. And I spoke about the show P which will actually discover and govern your complete information. It is, it is, it is very important to find out whether you have any sensitive information. You have to comply to the requirements of the different compliance rules like the GDPR which comes for, you know, the European countries or CCP California laws, right? So India is also coming up with the data privacy law and different countries are coming up with the data privacy law and we need to cater to those laws. And uh so we need to govern and our show purview comes handy uh for that particular requirement with that. Um I would uh conclude my session and I'll just see if there are any questions. Um Well, I see can these be used to focus output such as in retail where we can use and train A I to predict the inventory stuff. Absolutely, Sara, I think I just showed the one of the use case also the inventory management uh for our um Azure machine learning. We can definitely do that. So uh Aishwarya Puja, definitely we can use it any more questions anybody is having. No. OK, cool. Um Thank you everybody for attending my session and uh thank you uh for patiently listening. I would just share my linkedin profile. You can also take only Namrita Shirane and connect with me. Um Thank you for joining in.