 Hello, everyone. Thank you so much for making time for this webinar. Sanjay Thir, I will be presenting on machine learning and product management and hopefully I can take half an hour of this day and give you something, give some action it was to you about me. I'm Sanjay, I've done my BA from NSID Delhi and my MBA from ISP Azirabad. I have around 13, 14 years of professional experience. I've been in product for the last nine to 10 years of my life. I was software developer for four years, currently I'm a group product manager with booking.com based in Amsterdam and I take care of their technical platforms and unofficially I'm an accidental MBA who is looking to reclaim his tech roots and hence the reason for this workshop. So why are we here? First thing to give you that is commensurate to your time spent here. Secondly, to come straight on our lockdown lives, things are improving, but we have done this dream before we know this right so please be vigilant. Please take care of your loved ones. And finally I will give you a flavor on the machine learning flavors and product management. So my plan is to take you through the life cycle of data in a typically dataware organization that most of our tech organizations are like now and give you the flavors where machine learning is increasingly becoming so important and where are the value gaps that our product manager can actually intercede and add value to his org. So firstly, let's finalize the terminology. Okay, so what is artificial learning, what is machine learning, what is deep learning, right? Let's start from the super set. Artificial learning artificial intelligence system is a program that can sense, reason, act or adapt. Know where it is and it's supposed to be a silicon based program. The artificial intelligence aims towards singularity, what we call the singularity, some being which is sentient, which can take decisions on its own. Machine learning is a subset of that. It's actually algorithms that are working together to speed a data set to a let's say a silicon based system so that the data learns over time. And deep learning is even a smaller subset, basically multivariate heavy data sets. Machine learning and deep learning as terms are used interchangeably. You see the view on the right, there are a lot of such means that machine learning is kind of a freak in this world. But yeah, but it is statistics but with steroids. So I would like to call this statistics on steroids. Why are we actually getting involved? Why is machine learning such a rage right now? So training data loops has been around for generations. I think the first machine learning people was written in the 50s. So what has changed? Why is suddenly product management, product management businesses taking so much interest in actually machine learning? Firstly, data global. Everything in anything and the sun generates data. We capture data every day. When we sign in, we sign out, when we actually order food on our app, we actually anytime we open our mobile. There's tons and tons of data being generated, petabytes of data being generated by you and me as individuals every day. There was a time that Moore's law used to say that semiconductor, the amount of chips nearly double every X amount of years, it is long dead. Computing power is commoditized right now. It's growing with the leaps and bounds. So the divide that was that between computing power and the benefit of the analysis that we get from the computing power is far actually outstripped. That's why machine learning is a very pragmatic solution. And thirdly, collaborative open source. All of us are working on gate. All of us can see what either each other is working on. The machine algorithms, the data science algorithms are also improving by leaps and bounds, hence the extent, hence the keen interest in machine learning in product management. What I will now do is take you through the ecosystem of a typically data bear organization from the generation of data to actionable of the data. And then what we will do is we will pick up each of these data sets, pick up each of these layers, and we'll finalize and we'll actually try to discuss, okay, what are the problem areas in each of these layers, and where product management comes into the future. So basically what we start for is data sources. Data sources can be internal, they can be external, they can be clean, they can be rough data. But mostly any kind of a tech-aware system, any kind of a data heavy or tech system can be broken down into these five layers. So you have data sources, data sources feeding into aggregation and aggregation layer. So the first layer, which is data source layer is called the formative layer. You're forming databases, forming data sets here. The second layer is data aggregation layer. It can also be the data representation layer. There's a lot of data representation that also happens at the aggregation layer. The third layer is the declarative layer. There's the machine learning declarative layer or the tools layer. Or for the developers among us, there's the MLOps layer. And the fourth layer is the applicative layer, where we actually see the applications of the machine learning algorithm get into picture. Where the actual application for machine learning, deep science, etc. are written. And the fifth layer is the actionable. One thing you should definitely notice, there's a horizontal observability and monitoring layer going across these rules. Any tech-aware system worth its salt today has monitoring or observability. And of course, with the advent of machine learning into tech, you see a lot of tools, a lot of product gaps being in that space as well. So we will briefly discuss on those as well. Cool. So let's pick the first layer, which is the formative layer. So machine learning formative layer. What are the product gaps that you can think from you, you and me as a product manager? This area is pretty well now. We have had data sets for a very long time. I think the big data that is we are getting right now is something new, not so much new. But we have had data sets for a very long time. So basically this area can be divided into three parts. Storage or the storage, basically data now. So there's different evolution of databases. We started with no SQL, RDBMSs for the old people among us. And you see a lot of evolution in databases. You have no SQL database, you have SQL database, you have RDBMS, you have graph databases. In case you're looking at a user-centered organization. Product evolution in this area normally happens in co-areas. So there's a lot of advent on quantum TVs right now based on blockchain TVs. So if you're building let's say an old TP kind of a use case, you should definitely evaluate them. So basically what is happening here is the product focus, the product market fit, the product gap that we have in storage is pretty very niche. And this is an integral, let's say a SaaS product manager working in that particular area. You are optimizing on very narrow fronts. So from a let's say 30,000 point of view, there is a lot of activity here. But maybe not enough to generate mass interest. The second is the data transformation layer. Of course, wherever data is, you need a transformative layer. Data can be presentable, data cannot be presentable. In that sense, you will see a lot of ELT, ETL transformative solutions present here. Of course, this is again an area for SaaS PMs as well, enterprise PMs. Because you're building a lot of SaaS and enterprise applications here. And the third layer is wherever you have data, you will have somebody kind of a qualifying layer for data access. So in that access, in that issue, you will see a lot of data access solutions, which are ever present across access control. They are pretty vanilla, all of us know SSO, etc, etc, identity and authentication. And those kind of solutions, those kind of product are also in demand here. Moving on to the second layer, which is the aggregation and presentation there. What kind of product gaps do we see here? We divide this product area into four basic divisions. First one is aggregation, basically aggregation of data. This is where data becomes big data, multiple data sets coming together. As you know, this has been a very hot area. Aggregation solutions, evolved data solutions, data leak, etc, etc. There's a lot of demand for these kinds of solutions. So there is a ton of activity in this area from a product management point of view. The second division is catalog. Wherever you have data, multiple data sets together, aggregation of data sets, big hits need a proper catalog. This is a very hot area right now. You will see a lot of solutions that are trying to solve this pain point present in the industry, properly cataloged, big data. The third thing that you see is analytics and visualization. This has always been presented. Then you have data. How do you visualize that data? How do you do analysis on this data? It's ever been perennial product stream. It has always done well. It is always improving and you will always see new product gaps here. So product manager should always be aware of this. And finally, we have search. It's the most vanilla use case. You have data. You want to search something with data has been present from eons. A lot of things have been solved pretty well populated. Unless and until you are bringing something totally new, I don't see much product gap here. Finally, we get to the MRA declaratively. So now your data is formatted. The data is presentable and aggregated. Now before you actually can build a machine learning model on it or machine learning application on it. What do you need is that data should be really labeled is a term that is used a lot in the machine learning algorithm. And what we are looking for is a machine learning ops machine learning tool platform that enables me that enables me as a data practitioner to be able to deploy and do work on the machine learning models in the best manner. And this is a very hot area right now. Any conference, anywhere you look, any market that you look, you will see that. Okay, the applicative layer, which is the next layer is pretty hot. Yeah, we need a lot of input into it. But the value differentiator right now in the industry is declarative layer. How quickly can you enable my data practitioners, my machine learning practitioners to actually deploy and learn from these models is the difference between an organization that is doing very well in machine learning and an organization that is sending one machine learning. But again, let's try to break this down. We break this down into four parts. First one is data preparation. So even a data is ready. Data is aggregated and presentable. Before I can actually feed it into a machine learning model, a machine learning loop. I need the data to be well labeled. And labeled is the industry that is used as we discussed. It means, and it requires a perennial pain issue. It's a perennial pain point or ML practitioner. There's ton of growth activity happening there. There is a lot of product gaps that you and me as product practitioners can actually look and exploit in this area. The second one is model builders. So, even though you have machine learning practitioners, knowing how to actually, knowing how to form these blocks, actually a model builder is a backdrop of any machine learning platform. This is like the platform of a platform, platform square, if I can say that. And there's a ton of activity here. There's always ton of activity. Of course, this is the first thing that any machine learning platform works at. But there's a lot of problems that can still add value. The third one is ease of deployment tools. Anywhere, anywhere you have a platform layer, you will always have ease of deployment tools. Productivity tools that can actually help the customer's job better. And you will see a lot of activity here as well. You will see a ton of productivity tools borrowing heavily from personal productivity tools as well, et cetera. But this is a nice idea. This is a good activity here as well. And finally, a lot of machine learning platform, machine learning declarative tools are also dependent on use cases. So basically, let's say if you are, let's say a Uber working on self-driving cars, a lot of platforms, the components that are built there would be specific to the problem cases you are solving. So there's a certain amount of standardization that you can see that, but there's a certain amount of specificity as well. Of course, this is a niche area, but always active. As in when we have use cases, this area would always be active. And finally, we get into the applicative layer. Frankly, this area is very vast, evolving and cannot be divided as such. I cannot divide it into A, B, and C. But what I can say is what is the evolution of this area moving to work? The reason why this area is evolving so much is because we have better theoretical models. Our scientific community is joining our papers. You see a lot of theoretical models that are making their way into the practical world. And actually seeing a lot of applications. A lot of applications are self-driving synthesisers. Your voice to artificial intelligence synthesis is a prime example that I think a lot of the use cases are being solved on this. So better theoretical models, of course. And this is where a lot of PhDs, a lot of deep science practitioners are getting actively booed in the industry. And this is where, this is why this area is getting us a lot. And of course, better deployment strategies. They are bridging the man machine medium. Whatever we are looking at from theoretical point of view, we are getting much better in deploying that onto a software kind or a hardware kind of a mechanism. Rapid prototyping and hardware being commoditized have actually benefited this a lot. And of course, that's the reason there's a lot of code. But from a product management point of view, are you actually bringing something to the table? You are identifying a machine learning, let's say value gap, a gap in the market. But is it you who is driving this? Actually, back to the book. So from a product management point of view, you are not really trying this. This is purely being driven by scientific community. And finally, we come to the observability and monitoring here. Everything in a tech-aware system is monitored and observed. I hope you guys know the difference. And this is a prerequisite of a tech-aware system. Each product layer has its own unique monitoring capability challenges, including machine learning. So there are solutions. There are unique solutions that only work on machine learning observability. And then there are solutions that will work for everything. Definitely, there's a product graph here. Which you or me or one of the product management practitioners should keenly look into. So what are the trend lines basically for tech PMs with this ecosystem that we discussed? The strongest trend line is that data and ML Ops is not going anywhere. There's a ton of product market fit that is still to be exploited. This says that we have reached a critical mass. It will only improve. Data is the new goal, as you can say. Data is a new oil. And so anything related to data, anything data like is here to stay. And data ecosystems are expanding. But as you can see, if you look at, let's say, the amount of tools that are coming up, some amount of consolidation can always be foreseen. There are strong trends, strong, strong trend lines for this as well. What are the weak trend lines? Does applicatively need a PM? What would project manager do? Does a PM really bringing value to the applicatively? I am still on the fence, but I'm ready to be convinced otherwise. And of course, always as you're dealing with statistics, the answer is statistically variant. I hope I didn't bore many people. So that is me. Thank you so much. I hope that really added value to you on your Monday. And if anything else, please reach out. I'm always available. I'm always a LinkedIn messenger. Thank you.