MartechBrain | E18 – Tapan Barman on “Conversational AI”
In this episode, Rajesh Jain speaks to Tapan Barman, co-founder, mihup, about “Conversational AI: Re-imagining the Human-Machine Interface”
Rajesh Jain(host): Hi everyone, welcome to martech brain a net core initiative where we talk to some of the best brains in martech and in the tech space and delt deep into one topic. My guest today is Tapan Barman, co-founder of MIHUP. Tapan’s chosen topic for our conversation today is conversational AI reimagining the human machine interface, so human machine communications. So, welcome to martech brain Tapan.
Rajesh: So, tell us more about MIHUP ? What does it do ? And then we”ll get to the wider topic linked with that.
Tapan: Sure, so MIHUP is “May I Help You Please” , if you consider the full name of the company , so we considered MIHUP as a journey , which we have started in 2016 and we aim to build a company to create a greater impact for the society with some cool technology. So in the MIHUP what we are doing , we are building voice AI platform and we help enterprise developers to add voice interface for any device applications, so that’s how it works and we initially raised the investment from the access partners in 2016 and Subrato from excel wrote the first check to us and, i am the co-founder and i do have Biplab and Sandipan Chattopadhyay was the co-founder of this company. So we are about 50 plus sort of missionary team operating from Kolkata and Bangalore . So that’s about the company.
Rajesh: So what are the problems that MIHUP actually solves for companies?
Tapan: So look, when you are saying the voice AI platform you look at the problem imagine, you are calling a call center and you are navigating through a complex menu tree like for this you press one for this you press two. Now we are helping to remove the entire complex menu tree so that users can directly talk to a system or the machine where machines can answer most of the queries without human agents. So that’s the kind of problem we are trying to solve with the call center and also when you are calling, any calls or you hear, the call can be recorded for the quality and training purposes but the manual keyword team can hardly analyze one- two percent of the calls. So we are also helping enterprises to sort of analyze 100 percent of the interactions to get all kinds of insights which is helping them to improve sales collection and everything. So that is one part of the problem which we are trying to solve with the help of MIHUP. On the other side we are also building a virtual agent on the top of the platform for different kinds of IOT applications, say for example smart cars . So we recently deployed our virtual agent with TATA motors. They are using that virtual engine in their cars so that users can control everything inside the car just by talking, so that’s the kind of problem area where we are trying to build applications and trying to solve the problems.
Rajesh: So , in a way what you’re basically “saying it that for all of us speaking is the most natural interface rather than touching clicking and everything else that we have been doing all these years and you’re taking us back to the origins as it were how we would communicate, we are now communicating that way with machines”
Tapan: Absolutely , the way we come to take the example of India where 90 percent of the population don’t speak in English, and most of the applications or devices if you see the interface is pretty complicated , it is not very user-friendly . So what you have found most of the people are comfortable to speak their own native language rather than typing in the keyboards. That’s why I think voice interfaces are the most intuitive , which can help users to express their interests and desires in a more effective way. So that’s where it is going to be the change in the next era, where we’ll see a lot of voice devices happening in the market. So, I think these are the areas where wires can play key roles in transforming the way we are working with those digital devices.
Rajesh: In fact we’ve already seen in the last few years the rising popularity of devices like alexa, and now with smart tvs. You have the remotes where you can actually speak .Anyone who has tried to type using the tv remote, search terms etc it is very very painful, so voice becomes the natural way and even in car navigation systems like example , you talked about TATA motors, just speaking so put the ac on or play this or switch to this music station etc now become so much simpler rather than trying to navigate through the dashboards.
Tapan: And also those are absolutely some of the important and relevant use cases which are happening around us and also I think, if you look at the perspective of a farmer who just wants to understand which season is appropriate for say ABC for cultivation . So, how we can get the best help from some point and with the help of machines, because it’s very hard to help those in that scale of commerce with that kind of information. These kinds of virtual agents can be much more effective. And also think of virtual agents. Imagine a virtual engineer bank where you don’t have to travel to a branch just to inquire about the basic information. The way the JAN DHAN account holders are doing today in India in remote areas. So, we actually can enable them and empower them to equip all kinds of information. They want this kind of application which can run over an IVR system or the call center. So wide the numerous use cases possible and ultimately it will help a large set of people to empower them to get much more or get much more confidence on the information about the entire thing, so that’s where I think the change will happen in the coming days.
Rajesh: So, Tapan, why is it taking off now? So i think if we take the area of conversational AI. so what are the building blocks which are helping conversational AI really rise onto the forefront in this. And I also want to take up this element of we’ve talked about the conversation part and we also want to discuss in the AI part where does the AI come into the whole system?
Tapan: Correct, so Rajesh in conversation also we consider conversation in four different categories. It can be human to machine machine to machine machine to human and human to command, so these are four types of conversation which happen. And what is AI? AI is basically, you train a machine as a computer with certain knowledge with certain data which actually can help a machine or system to drive a conversation or drive or drive certain instructions, that is called artificial intelligence. Say for example, you are a call center agent and you are actually talking to a customer. Now, imagine while you are talking to a customer you are getting some sort of guidance from the system itself but the system is telling you and going first you , you speak slowly . you are talking nonsense you should talk in the right way. So that kind of instructions can come from the machines. when the machine is artificially intelligent with some pre – loaded data. Now how we get those data, actually we get from the users uses the way we talk and the way we communicate from all kinds of user behavioral data is actually we are using to train machines and we are using the best to train machines to train the worst at the bottom. So that’s how artificial intelligence works. Now why it is taking up now, because there are so many important things happening. First of all, I think the most important piece is , if you have to train a system we call it machine learning so you have to train your machine with more data. It requires lots of computers and earlier it was very very expensive. You see the share market price of the NVIDIA which is now going up very much right so that itself proof that NVIDIA is a company where they are doing pretty well and the devices that they are producing which we use for this computation is actually available at some reasonable cost, so compute cost is available at reasonable price, that’s what is driving this a lot of machine learning activities across the world and in this domain not just the voice multiple activities are happening. So I would say the compute cost is lower and second is now we are at a point where we are sitting with lot of data, enough data which is like unused data so down now the enterprise and the companies or I would say the researchers the developers they are trying to find out how much insights and how much insights you can get from those data and what’s the best way to do that? And it to get next set of your customer, say for example flipkart ,one customers from tier 2 tier 3 repeatable customers not just the first time customer how it is possible ?Second, for example, I just want to improve customer service or I just want to know my customer profile in a much better way than how it is possible! Third, for example, I live in a city where my nearest city scan system or the facility available is far away from 50 kilometers. Is it possible to get a city scan system or facility here in my town and since I don’t have a trained operator to maintain that? Can you use that with AI , so these kinds of requirements actually drive those factors in this space.
Rajesh: So, Tapan, how did you get interested in this area? Tell us a little about your background and how did this whole conversational AI get to the forefront in your life?
Tapan: So about my background, I come from a very , I would say village. It is in tier three cities. I would say it’s very close to the Assam and Bangladesh border and my native is Rajbongshi , and the language I speak my native is Rajbongshi which is not even Bengali not in English and which is not even part of this 22 scheduled language of India. But there are three million people who speak in Rajbongshi. So for me I feel language is ,speaking English is sort of a barrier for me. I’ve learned how to speak English to get into my profession. So I also believe if you don’t speak , if you don’t have the opportunity to speak your own native language you lack confidence. So that’s why, I was always interested to know about how to solve this problem. Initially, maybe some translator engine I’ll speak in Bengali and get transferred to English it is relevant for multiple languages and. so backward wise. I graduated in computer science in 2006. I was working in a public sector company for two years and started my own company in 2008 but was doing reselling business with the consulting business in the company. But I always wanted to build this application to solve this problem and finally I got the opportunity in 2014. Where i got introduced to Sandipan at the city of Justdial who was having 10 years experience of building or working with the voice. And also thankfully people are binned. Finally we got into this problem space and started building those applications around the problem, so that’s how we got started and got attracted to this problem.
Rajesh: Fascinating story and I think this problem that you talked about breaking the language barrier for communication I think is very very critical especially in India, where we have so many so many different languages and even variations in those languages. I’m assuming one of the things that you have to do when you’re training the AI system is of course you need a lot of data but you also need to train it in the domain understanding.
Tapan: Correct yes.
Rajesh: And so in a car, so it’s not an open-ended problem ,and that is I think very important to where MIHUP plays a role right?
Tapan: Absolutely, I think it is relevant. Say for example when you’re talking to a person and we talk in a particular topic or context and if you can follow that conversation, he hardly remembers that 200 words in that entire conversation which was very very useful. So we also believe since this is machine learning in AI and which gets improved with enough data and the more you use AI it will get you to improve, that’s how it works. So i think you always debate about the accuracy or about the AI system or the other particular system. So we always said it has to be accurate within the context, is it accurate enough for that particular context. The system needs to understand everything it is not even required. So that’s why i think, if we want to solve this problem so we have to pick the problem or the context and then we should try to attend or try to solve the problem rather than solving the problem for anything it doesn’t make any sense for us so whenever you’re trying to our approach to to build the solution of the platform is that, use case is the primary you focus with use case and what problem you are trying to solve with the help of this particular AI or the conventional thing and then you try to build the model accordingly then it will work perfectly for the use case.
Rajesh: And like, if we go back to the TATA motors example are you also confident that even if the languages spoken are different , you can handle those scenarios or is it limited to say only English or Hindi or i’m assuming there’s a translation part of it which is also involved before we feed it to the AI engine.
Tapan: Correct, so absolutely we are confident so far. We have rolled out four language. We support Hindi English Tamil and Bengali, and we are also rolling out another six language so we do have a plan to support in language by next year. And we are also doing not just only for TATA private vehicles but also for the commercial trucks where the user is not very sophisticated, if you can imagine that dialects and accent could be hugely different. Now the way we have to build the platform and that the way we have trained it is capable to understand multiple dialect and accents within that context and that’s why we don’t focus on on the takes output, otherwise we generally focus on meaning so we call it speech to meaning engine, where from the dialect itself can get actual intent from that part. So that’s why I think. since we we focus on the speech to meaning part that’s why handling multiple language within the context is not a challenge but again as I said AI can be improved with more learning and more data. That’s why we do have a sort of feedback loop in the entire system where we collect those data through a mobile company, an app which goes to the cloud and comes back with the improved version that’s how it works. But generally the entire system works in that, in the car or the trucks is without internet so you don’t need connectivity.
Rajesh: Right, so in a way the power of this. Really is that if we look ahead few years this whole human machine interface where we’ll be able to speak to the gadgets in our homes to the devices that we have to possibly air conditioning units the tvs the washing machines the refrigerator house, so all of these devices which are becoming more and more complex because there’s a lot more digital elements which are coming into these devices and what’s also therefore changing is the way we communicate so rather than going through keypads and layers of interfaces we can just speak and get the get the job done.
Tapan: Absolutely, so entire thing, the user interface will get through a transformation and it will be like first device, so in the future there will be sort of no key recently apple they have sort of launched a prototype of that kind of future device and where there is no keyboard it’s just a voice with siri so that you can operate the entire machine with the voice. So that kind of future we can expect from the domain, but i think in India i still believe though the alexa of course is bringing new trends they have created a new opportunity for a lot of companies but for India alexa is still not relevant “i believe” because most of the Indians they live in 700 square feet of house where alexa is much much more bigger. I think from the Indian context and perspective we are still left to install a lot of confidence to Indian, so that they can effectively use mobile phones at least in the proper way and they have to believe with the help of the tools which are very very powerful. Smartphones or maybe normal phones, you can do a lot of things. I think that’s where the confidence is not there because I think most of them think this is something not for me. so that fear has to go first , so that they should get connected to the digital ecosystem, which you’re building and that digital platform the health everything is connected to. So that’s why I think that transformation has to happen in the mindset first in India and the devices for India, different kinds of devices are required, not the Alexa .
Rajesh: Correct, so one question really on how marketers can leverage this sort of revolution that’s coming in conversational AI , how can marketers make use of this technology?
Tapan: Oh many multiple ways , marketers can effectively use this particular application in several ways. I can give you a couple of examples where we have sort of helped multiple companies in our own way , say for example we are working with the edutech company, they are trying to onboard new students on their platform. Now they want to analyze the counselor performance, when the counselor is being counseled as students whether the counseling was appropriate or not? How effective was it? so because that’s happening in skill and humans can really assess each of those counseling sessions. Now they are using AI platforms where we are analyzing all the counseling every day and we are giving recommendations to individuals to team or to the companies, where they need to improve so what’s the best practice? What’s the thumb rule? What are the best performers doing? How can we sort out which clone is best for the bottom? So this is one of the examples, you could also use . We’re all also working with insurance companies, they are using this platform to improve their sales because they are saying look I do have two thousand callers every day, they are calling customers but my conversion is very very low. So how can we help me? So we are analyzing those in. The entire call and we are telling look this 10 salesperson is your best and how they are because this is the pitch that’s how they’re talking to their customers and this script has to be used and we are using that script to train the system for the machine. And the machine generates feedback for the rest of the agents.
Rajesh: All the other agents excellent
Tapan: That’s how marketers are getting benefits out of this product for the system
Rajesh: So, looking at the broader impact of this on society Tapan, what do you see as the impact if you look ahead? What’s the impact of conversational AI really on society in general?
Tapan: Yeah, i strongly believe as I mentioned in the change in the mindset I think i can give you a lot of examples still in our society especially in India we consider any computer or the devices the machines is an evil factor and because of lack of education of the literacy a literacy of i think that about multiple things, we still consider is not great to be in the social media platform because that is something not allowed in our society in some areas. And as I said we still fear the machines. We have a lot of people who don’t know how to use atm machines still today . So I think fear of machines is something that is the biggest factor which is driving a lack of confidence in us. I think if we want to transform or change your society the entire citizen has to be empowered and we’ll have to install a lot of confidence and that does the entire citizen has to become smart enough, so that they can make better and informed decisions. And that is only possible when you empower them with a tool which is very easy to use and with that tool they can actually access to anything and that’s where boys can play a very very effective role.
Rajesh: And and I presume it would be Mihup’s vision also!
Tapan: Absolutely, we just want to be the trusted factor like as i said talby, MIHUP we want to be the one of the trusted factor for everyone and they should think of MIHUP while they’re using voice to access anything any machines or the applications or anything that’s what our aspiration is.
Rajesh: I think, it’s a great view into the future I think basically if I were to summarize two or three of the key points that you’re saying is that we are going to see the rise of conversational interfaces, where we will be able to communicate just by speaking onto devices. The human and machine interface will keep improving the more the data. That is there the better the interface can be and even in the example that you gave in call centers or in conversations you can actually now remove the inefficiencies and the differences which are there between sales. People you can learn from the best and the others can be trained and even guided possibly in real time in conversations that are taking place.
Rajesh: So I think we are opening up a whole new fascinating world of conversational AI. So that was a wonderful introduction to the area of Tapan and we will come to the last part, the fun part, the five ones so quick sort of short answers. One future tech trend that excites you and other than of course conversational AI?
Tapan: Okay, so I think other than thinking of conventional AI today i am fond of two things and i’m truly excited about the potential. One is of course the way we are leveraging internet for the education and so MIHUP is working with a with multiple edutech companies and i can see today from a place where probably there is no good teacher may be available good in the sense of in this in the context of getting better indicating the foundation. I can think of accessing a teacher based out of the UK for the education of my child. I think that is something fascinating and it’s going to happen crazy. I think it’s good for the company, for the country and the society. And second fascinating thing I think on the health care side I think the way we are thinking of transforming the entire treatment. I follow Siddhartha Mukherjee and he talks about re-engineering cells, how we can re-engineer the genes and cells, instead of using pills that kind of , I think innovation is truly fascinating and really changes the way we are thinking about the treatment in the future.
Rajesh: Great insights, one tech or trend that has disappointed you?
Tapan: I would say the way we today use social media platforms to influence decisions of individuals, I think that is not correct . Recently if you have seen the advertisement of the whitehead junior the way they are sort of an advertising if you learn this code you’ll get a job in Google ,that kind of advertisement influencing is very bad and that trend has to be stopped..
Rajesh: Very good point. One good book recommendation?
Tapan: So, I think the great book is “Zero To One” by Peter Thiel. I have learned a lot from that book and yeah so i’d recommend everyone who is thinking of starting a company.
Rajesh: And one good person podcast to follow?
Tapan: I would actually suggest two people probably again. I follow Elon Musk very closely , because he talks about impossible things and makes them possible that are truly inspiring, and again the Siddhartha Mukherjee , the way of innovating the entire health care. I would say the thinking at least.
Rajesh: Absolutely. I think both have amazing visionaries. And final question. one key driving belief in your life?
Tapan: I always believed in one quote from Swami Vivekanand , that was a gift from my mother when i was in class five, the book is the power capsules, and there is a quote which says “the success is equal to purity plus passions plus perseverance”. I truly believe in that and I have seen a lot of people in my family and my colleagues and everything. If you truly stand for those three things it will be successful. And in life success is something the changing post it’s nothing fixed so those three cons factors have to be there for any sort of achievement here.
Rajesh: Great great suggestions.
Tapan: Thank you very much, it was a wonderful conversation with you. I wish you a good future for the world of conversational AI. I think we’re just seeing the starting point of the power of the internet and AI and technology with the following costs of computing storage. All of that is now coming together to really open up new opportunities going forward especially in the human and machine interface thank you very much, for the great chat with you Tapan.
Tapan: Thank you so much thank you.
Rajesh: Thank you everyone, for watching and listening. I will be back with another edition of martech brain next week.