Nirman Dave - Obviously_ai - Spotlight-MP3 for Audio Podcast...

Sun, 4/26 5:40PM • 29:57

SUMMARY KEYWORDS

machine learning, people, predictions, data, typically, customers, algorithm, tools, code, piece, machine, experiments, users, key, build, questions, learning, predict, run, business

SPEAKERS

Nirman Dave, Ben Tossell

Ben Tossell  00:00

Nirman Dave  00:25

yeah, for sure. So, um, you know, my name is near mine. I think my background is in machine learning and economics. And a couple of years back, I was actually working at a startup in San Francisco and we get about 70 ish people at the time. And I was the only data science person at the company. So what used to happen was a lot of non tech, business analyst and marketing analysts these guys would come up to me and they would ask me questions around data. So something as simple as Tell me who are top hundred question To do something more complex, like tell me which customers are likely to cancel. And every time they asked me this question, I had to go out and build a small machine learning model that would do the process and give them the results back. And then they would come up to me with follow up questions. And the problem was that it kind of took me away from the job I was actually supposed to be doing. And that's when the light bulb went off, which was like, why is it that these non tech business users, they're the ones asking questions, but the tools are made for technical people like myself? So can we build a tool that's incredibly easy to use, doesn't require a background in programming and can be used by anyone to do machine learning predictions without writing code. So that's where the idea really started. And today, the company is called obviously AI. And we are a no code data science platform for business users. So typically, all you have to do is plug in your data source. So upload your data, upload your CSV file with your customers and info on it. And then you get a Google like search bar to ask questions in it. So you can literally say, tell me which customers are likely to buy again in the next one month. And our technology will automatically understand what you're asking for in the right data and give you the prediction result back by building a machine learning model on the fly. So that's basically what the tool does.

Ben Tossell  02:27

Awesome. So how, I mean, how does it work? Obviously, I'm not expecting you to tell us all the all the details, but sort of as a high level, is it a bunch of machine learning components, if you call them that, that sort of sit together in one place, and it just the search sort of runs? Run a few of those, or how does that work?

Nirman Dave  02:47

Yeah, great, great question. So typically, for a good machine learning algorithm to work. One of the most important thing is that the algorithm needs to have a cup, a series of settings Right, and once you get the settings, right, it will do the job very well. So think of it as kind of like having multiple different knobs, and you got to just kind of change each knob and you got to tweak each knob. And until you get the positioning, right, the machine hasn't worked, right. So think of it that way. Um, so technically, the way obviously AI works is once you put in your data, we have a series of eight different algorithms. And it takes those eight different algorithms and runs all of them with hundreds of different settings, right? And then it picks the one that gives the highest accuracy. So typically, what it's doing is it's taking a vanilla algorithm, customizing it for your data set, and then using that algorithm to make a prediction.

Ben Tossell  03:47

Awesome. And then what what do you see people do with those predictions? So they tend to that needs their other tools to do stuff, or is it more just yeah I suppose it's more of a. Okay, here's the data now go off and like, put some actionable steps in place with with your customers and things like that. Yeah.

Nirman Dave  04:12

Great question again, I think so there's three parts to it. The first part often is people coming in and bringing in the data. And that's kind of the kind of structure the data to figure out what they want to put in the data set, stuff like that. So the second part is where they plug put the data into our system, and say, here's my data set, give me a prediction for x, right, and then our system will automatically build the algorithm, give them the prediction back. And the third step is where they take action on. On the results, when we primarily focus on is the second step, which is once you once you bring in the data, we'll give you the algorithm that gives the prediction and then you can do actions on the predictions itself. So that's kind of been Our focus today. Technically, and we're now trying to figure out ways in which we can go into different avenues to really double down on the entire process itself. But typically, our focus has been this one in terms of who uses it, and what do they use it for. We've had a little over 1000 users and obviously today, and these people come from very different industries. Typically, what we see is industries like marketing industries, like retail industries, like e commerce, these are the industries that really shine up where what they do is they have specific customer data, that is individual data for each customer's and to bring that to understand their customers better. So often, they predict churn which is who is likely to cancel, they predict Net Promoter Score, a likelihood of conversion and likelihood of buying again. So these are the typical kind of things that we see often come up, but also to be very creative also. Seen micro loans startups, which kind of use us to predict who is likely to pay the bills on time? We've seen? Yeah, we've seen the multiple different things, insurance startups that are predicting which healthcare plan people are going to go on. So people can be as creative as they want. But typically, this is kind of where the audience generally comes from.

Ben Tossell  06:25

Yeah. So I suppose you you sort of assumed that people were going to use this for their current customers, and then helping them keep those customers but actually, some people are using this for prospective customers and figuring out are they the right customers for us? Can we help service them and things like that? So how, how does that fit in? Is it just a guess someone would have a questionnaire of some sort and then use that to plug their data through, obviously AI and then they get themselves?

Nirman Dave  06:53

Yeah, exactly. So typically, what we see is, first of all people bring in historical data, right? So this is data sets about their customers that have already converted, for example. So they're bringing this historical data, they make a prediction on obviously AI. And after they're done with that, they set up a Google form, which their new customers have to fill out. And this Google form has questions that are similar to the ones in historical data set. Like, you know, what city are you from? What's your use case that data down? Yeah. And then some of them also use Zapier to connect the outcome of the Google spreadsheet or the Google form to our API and say, Okay, now this is the new person that came in, and these are the answers that they've given. Given these answers, what's the likelihood of conversion score for this person? So that's the kind of very interesting workflow that we've seen people do more often is they build a prediction and obviously, I tend to build out a questionnaire form for the new customers. And as new customers answered that question for every question, have a sense of the likelihood of conversion.

Ben Tossell  08:03

Oh, interesting. Yeah, maybe we should do that with make pat down and see what, what are the things people are looking to build and whether they're likely to be a good customer for MiG pad and make matters like, USA? And I guess we should do a tutorial as well. Absolutely. And so with machine learning, I feel like it's, it's one of those worlds that I know very little about. And I would butcher it if I was ever going to sort of talk about it too much. So what do you think? The note like? Why do you think no code and machine learning from sort of pairing up together? What what would that do for broadening the kinds of jobs people do in machine learning?

Nirman Dave  08:43

So I think, you know, one of the typical things is, but if you look at the scenario today in the world, the biggest thing is that machine learning is kind of it is very complicated, right? Even though it's kind of thrown around everywhere. Everyone claims to be doing it in some way or the other. At the core, building a teaching a computer, about real life data can be very complicated. And so typically what happens is mostly when you are a non tech person at a company, and you want to make a data driven decision, right, you want to make decisions for the business that are really well, you want to make decisions for your customers that can be highly impactful. And you often use data for that, right? So what we typically see is that these business users often kind of care about getting to the decision, and their professional skill set isn't being creative with the kind of decisions they can make. And because machine learning is so inaccessible, because it is so complicated to do that often have to depend on other people to get the questions answered. So one of the key kind of change in the world that we'll likely see with no coding machine learning coming in picture is enable these business users to find Focus on what they are doing really well, which is being creative, rather than waiting on other people to build an algorithm, kind of get into the weeds of how the algorithm works, stuff like that. You know, Funny enough, technically, nowadays, we've seen some business analyst trying to learn machine learning themselves when they have no background in programming or math. And it can get very complicated very fast. And that becomes a little bit of challenging moving forward. So that's one of the key changes that you're going to see on the creative side on the business side, which allows people to be more creative on the machine learning side itself. Another key changes that you'll see is a lot of engineers that are into machine learning. When they're typically hired, they're hired to work on some really exciting projects. But often, over time, they see 80% of their time being spent to kind of find the data, clean the data on some of the questions of other people and very little time on things that they were actually hired to do. So one of the key things this is going to help open up, it's going to open up avenues for existing data scientists to focus on work that they actually want to do things like building a face recognition algorithm, building a recommendation system, and while it leaves the heavy lifting of the algorithm itself, on the tool, so that's the kind of benefit that we see on both cases. And that's the kind of change that we

Ben Tossell  11:29

plan to bring in the world. Yeah, I think it's like, it's so interesting to see the similarities between different tools and note the no code space in general in that. I think we're all trying to remove the boring work, right? We're all trying to remove the clunky repetitive tasks that anyone has to do. And like I said, there's people go into a job to do the most interesting parts of that job. People don't necessarily go and be an engineer to have to spin up a like the basic login system every time and like all these basic things they want to do the cool things that push

Nirman Dave  12:03

like that

Ben Tossell  12:04

20% of the the stuff at the end. So I think there's a lot, there's a lot of a lot of things to be said about, like how these tools just push you and get you that 80% of the way there a lot quicker, a lot easier so that you can focus on being more creative. But yeah, I think we need to, like, shout out about this a lot more to a lot more people and do it that way.

Nirman Dave  12:28

Yeah, exactly. And think of it this way. And I typically what we see on an average sized company, is 80% of the employees are employees that are doing kind of everyday tasks, right. So these are business users making business decisions, analysts trying to figure out data and you know, business models and trying to really figure out how it works. And then there's a 10% which is kind of the leadership of the business and the rest. 10% is often kind of on the research, team and research side of things. And what we see in the research side of things is that often in the company only 10%, which the research group often does, quote, unquote, cool stuff often runs, quote, unquote, new experiments, right? Everyone else is kind of doing the same thing over and over again. So one of the key things, this kind of change brings in, and this is just a key thing for no code, along with no code machine learning. Imagine if everyone in your company had the freedom to run more experiments. Imagine if everyone in your company had the freedom to be able to be more creative. And do it faster, right. So this is kind of at the no code tools come in, where it helps you automate, and helps you do the kind of like, mundane stuff faster without you being too involved in the code. And so you can focus on being more creative. So that's the kind of thought process that we come from is kind of enabling that 80% chance of the company to run more experiments to be more creative and not worry about the coding side of things.

Ben Tossell  14:06

Yeah, definitely. Yeah, I'm just so funny to see that it is all the same, the same sort of walking with with the tools. It's awesome that everyone's inadvertently pushing for the same things. And I think we've all seen our fair share of and creative work that I do. do our best to push that forward. Yeah. What are some things in machine learning that are a lot less complicated than most people think?

Nirman Dave  14:35

machine learning a lot more complicated and most people think I'm typically a lot of people believe that. So one of the classic examples is, I want to predict if if this image is a cat or a dog, for example, and this is this is just machine learning in general. And often a lot of people believe that okay, I can put in a neural network, and that's going to do the job. And because neural networks are so commonly referred to, a lot of people believe that the algorithm is the one that does the job well, right. So picking the right algorithm is half the battle. In reality, what happens is that it's, it's actually the algorithm, once you pick the algorithm that has to be a human being has to sit down and manually kind of tweak every setting of the algorithm. So think of an algorithm setting as kind of driving the plane, right? When you're driving a plane, or flying a plane, you basically have so many different buttons and so many different knobs and you just have to get the right combination for the plane to be at a steady speed. So that's exactly what you're looking to do. When you do machine learning. There's so many settings, there's so many different knobs, and you have to just get the right combination for it to work really well. And that's the that's the piece that a lot of people Don't really often see on surface is because that's the thing that an engineer has to sit down and keep doing manually. on surfaces, oh, there's a neural network that predicted x. But really, if you look at a neural network, it's about getting the right combination of the settings. And so that's something that we we often don't see a lot of people getting into the desktop. So one of the key challenges of no code machine learning for us when we started was how do we automate that process? How do we automate the process of getting the right setting combinations without the user being too involved? And that's kind of one of the key changes that we bringing in that enables people to know code machine learning.

Ben Tossell  16:48

Yeah. And as part of that, just the product or is it more of a documentation and indications piece? Oh,

Nirman Dave  16:55

yeah, I think it's both. So there's a higher level documentation and education people That he can kind of see around on the internet. But there's also kind of the product piece where we have tweaked certain things that enable us to do it much faster than otherwise. So typically some process that would take a couple of hours if you were to do it automatically. We could do it in less than a minute. And basically, that's the product piece where it enables us to build a differentiation, then, just the overall education.

Ben Tossell  17:31

Awesome. Yeah, we need to play around a lot more with these things. And looking forward to doing that what you wrote, you wrote a post recently that was was the five no code, predictions, data science predictions for the 2020s. Can you walk us through some of those?

Nirman Dave  17:50

Yeah, for sure. So let me bring that up and kind of go into that. So you were talking about the the post We did, I'd say around in January 20 minutes. Got it. So this is something that typically, we would looking at just generally, around how no code is affecting different industries. And one of the key things that we wanted to look at was a, like algorithms related to machine learning. And we wanted to look at kind of just no code, you know, core tools that are used by engineers in the space. So one of the first thing that we mentioned is, the general public will crack down on the bias algorithms, which means that typically, what we see is that so many machine learning algorithms out there that are biased because the data wasn't well structured, the data input on that machine learning algorithm wasn't bad done. So the output of the algorithm is quite biased. So one of the key things that we're doing Through no quarter machine learning and through no code, because now people have more access to the data and more access to tools that enable you to change and manipulate the data will start seeing that public will be able to crack down on these biases. Because there is no code machine learning tools available. Users can understand what machine learning means users can understand and play around with the data. And with that combination, they'll be able to figure out how to decrease bias in billing models. So say, for example, something like obviously, the AI, let's say you upload a data set of customers that are only male, and you upload that, upload the customer data, and obviously, I will quickly start seeing bias and as a non tech user, you'll be able to recognize it, and you'll be able to change it and say hey, you know, this is what I should be doing in order to avoid it. So that's one of the key changes that we think when you think of no code pool is bringing into A picture for machine learning. The second thing is roles and responsibilities change, which is something that I touched on briefly, which was now that you as a, quote unquote business analyst has the ability to do no code machine learning. However, how is your role going to change? Right? So now your role is going to be more on the creative side, what we typically believe is we will see roles such as user scientists coming in more often, which is, which is a, which is basically an individual that focuses on experimenting with user personas to get to a better output. So for example, a user scientist would be someone that would say, hey, I want to experiment with this persona and see their likelihood of cancel. I want to experiment with this persona and see their likelihood of buying again, dotted under that. So this is a user centric person that runs experiments for personas. And they'll be able to do that because of no code. Machine Learning Tools that's on their toolkit. And they will be able to do that fast. Typically, you cannot do that today because you, if for every experiment you have run, you have to wait for a couple of months on an engineer who will give you the results and getting follow up answers to that is going to be more difficult. So that's one of the key things that we will see change as well. And then one of the key things we hope to see as well is the third point around SQL query is disappearing. which ties into the first piece of often people have a lot of data, especially on the databases, but they have to write complex code to extract that data. Typically, with natural language bringing coming into machine learning, that piece will be automated. So that's one of the key process and the the next two things are more around how the process of machine learning will become. So think about it this way today. Machine learning is very non collaborative, meaning it's something where you have an idea you want to predict. You go to a data scientist, they make a prediction, they give you the results back. And then you present it to the team that takes overall about a month of it's a long project, it takes about a month. But imagine if you're doing a KPI meeting in your company. And in the meeting, you have this big screen and on the screen, you pull up the software, and you say, Okay, let's run a couple of predictions right now. And let's see, what are the typical type of actions we want to take. That's the kind of features that we'll see with machine learning because the process will become more collaborative, it will become like, think of it as kind of, you know, any kind of BI tool that you've seen out there, which is a bunch of graphs, a bunch of data visualization that you can just pull up on the screen. Similarly, we'll see machine learning predictions being able to be something that anyone can do while In the meeting, and talk about it and kind of take actions on it. So we'll see a lot of collaboration piece coming in, as well. So these are the kind of predictions that we've made based on how we think the industry with no code machinery is going to change.

Ben Tossell  23:14

Yeah, no, it's definitely interesting. I think, do you think that sort of the introduction of these no code tools that help you move further? And he talked about the roles and responsibilities changing? Do you think that roles maybe in just data science, or even in a lot more specialized industries even become even more specialized so that the bar is almost increasing all the time, maybe like in leaps and bounds because a lot of people could access? Like I could technically now go to obviously AI and start running some, like machine learning data predictions, and I could do like a base layer of that you think ran that sort of pushes people to become way more special. In that sense,

Nirman Dave  24:02

exactly, exactly. It's like think about it this way. For example, if a data scientist who spends about 80% of their time trying to answer questions for other people now doesn't have to do that, because the non tech business user can answer their own questions. It frees up this person to be able to get specialized in one particular aspect of machine learning. So that's one way to think of it. Second way to think of it is the business user themselves can now specialize in running experiments for a very specific type of industry for a very specific type of persona, and can know that industry really well, based on the experiments that they run. So exactly, as I mentioned, um, it'll open up gates for people to be more specialized.

Ben Tossell  24:48

Yeah, definitely. Yeah, I think I definitely agree. So what do you think is next for obviously, oh, where were we gonna see you You go in

Nirman Dave  25:01

So I think one of the first key challenge is where we are today, in terms of where we want to be next kind of moving forward is a Our vision is to make data science effortless for everyone. And in part of that vision, one of the key things we want to do is we want to become the tool that enables anyone to literally just pull it up on the screen and run predictions while they're in a meeting, and ask for questions and kind of become collaborative. So that collaboration piece is something that we're really shooting for, where we want to enable the entire team to be able to come in, ask more questions and run more predictions as if it's just another as if it's like just making a graph or visualizing some data set. So that's the kind of ease of machine learning that we want to bring in. To do that, we'll take a couple of steps, right, so we're doing a couple of steps on the product. So one of the key things that we have Focus on four now is the three pieces. As I mentioned, the first one was the data collection and processing piece. Second is the prediction. And the third one is what we call the prescription piece, which is what tells you what action you should take. So now that we are perfecting the second piece, which is the prediction piece, once we hit that, we'll be looking to advance into the to other segments, which is kind of data collection piece. And one of the things that we've already done for that is build out something called a data store, which is like an app store for data sets. And the second thing we're going to be doing with something like prescriptive piece is we'll now be moving forward and connecting with other tools like Zapier and bubble that you kind of people that pretty much used to seeing to say okay, if this prediction happens, automatically take this action. So those are the kinds of A process that we have moving forward that eventually what we want to do is we're going to make the entire process end to end for data science incredibly easy to use.

Ben Tossell  27:10

Oh, yeah, that'd be that'd be amazing to see this sort of actions taken, do you think there's, I mean, the only piece of data science site that I'm quite aware of is kaggle. And they have like their they've got datasets on there. So I completely get that there's like an app store component would be really, really interesting. Is there going to be any challenges and things like that, that you're going to start hosting? Or is there anything like that the community aspect where we can, because I see a lot of companies like webflow, for example, who have built their own tool, and then they almost almost don't even realize the power of it until it's in the hands of the community and then the community push it so far, then they see different use cases and all sorts of things like that. So I wonder if, if that's something you're you're thinking about doing and using the power of the community?

Nirman Dave  27:58

Absolutely. And actually I think one of the key things we've already started doing is engaging people over our Twitter community, which is very vocal and very connected in the space. So that's kind of where we are starting at. But eventually, as we go down the line, we want to kind of build up this community where people are asking about different predictions and making, they're sharing the insights that they're getting, kind of collaborating on it. So that's kind of the goal that we'll be going through. But I think for now, the community starting at twitter. So for example, very recently, we asked everyone on Twitter, what they would like to see in a playbook that we're writing in a playbook of machine learning what what is it that they would like to see? And we've got a really good response in terms of people kind of giving suggestions that are giving ideas. And what it really tells us is, you know, whatever we do from now on will be driven by the user a lot and the That's something that we're shooting to do. So our starting point is Twitter. And obviously, we'll expand from there.

Ben Tossell  29:06

Awesome. Well, we're looking forward to seeing that. Is there anything else you'd like to promote to the community today?

Nirman Dave  29:12

Um, yeah, I mean, so would love to again, just give a shout out to our Twitter and, and our website, which is obviously.ai, where you can get all the information about what the tool is, and just sign up and get started.

Ben Tossell  29:29

Awesome. Thanks so much for your time today. We're looking forward to work with you in the future and and see what other people can do with with machine learning at their fingertips.

Nirman Dave  29:38

Likewise, well, thanks so much, man.

Ben Tossell  29:40

Really appreciate it. Awesome. Thanks a lot. Cheers. Thanks so much for listening. You can find us online at maker pad.co or on Twitter at make that we'd love to hear if you enjoyed this episode, and what we do next.

‍