sftext.org: Alexy Khrabrov interviews Babak Rasolzadeh

hello everybody I'm Alexa crabber of the hello everybody I'm Alexa crabber of the organizers SF text meet up and here organizers SF text meet up and here we're on location meltwater a very happy we're on location meltwater a very happy to be at melt water for the first time to be at melt water for the first time and here with us we have Bob across all and here with us we have Bob across all today who is a director of data science today who is a director of data science it melt water and Bob walk is going to it melt water and Bob walk is going to talk today about NOP technology said not talk today about NOP technology said not water but we're going to ask Baba hear water but we're going to ask Baba hear about his own interest so how you know about his own interest so how you know did you come to do data science at did you come to do data science at meltwater what was your career meltwater what was your career trajectory before ya know you are in trajectory before ya know you are in different startups in different parts of different startups in different parts of the world you know if you can kind of the world you know if you can kind of give us an outline of you interest and give us an outline of you interest and what attracted you to this space sure what attracted you to this space sure absolutely very happy to have you here absolutely very happy to have you here and have have the meetup arranged here and have have the meetup arranged here at meltwater this time my background and at meltwater this time my background and in this field is is actually not in text in this field is is actually not in text analysis or NOP my background is analysis or NOP my background is actually from computer vision and robot actually from computer vision and robot robotic vision to be precise I I did my robotic vision to be precise I I did my PhD in that subject and doing those PhD in that subject and doing those those five six very forming years this those five six very forming years this was back in 2005-2006 I i worked on was back in 2005-2006 I i worked on everything from autonomous vehicles to everything from autonomous vehicles to self-driving cars and technologies for self-driving cars and technologies for pedestrian detection street sign pedestrian detection street sign detection to technologies for humanoid detection to technologies for humanoid robots to interact with their immediate robots to interact with their immediate environment to actually be able to environment to actually be able to manipulate the objects they see to manipulate the objects they see to recognize the objects or even be able to recognize the objects or even be able to manipulate objects that they haven't manipulate objects that they haven't recognized but they can identify the recognized but they can identify the shape so a bit of a diverse background shape so a bit of a diverse background in that sense i was was ever good doing in that sense i was was ever good doing my PhD yet focusing as my my advisor my PhD yet focusing as my my advisor used to say my professor that that you used to say my professor that that you know i need to focus on one thing i know i need to focus on one thing i heard focus focus focus for my advisor heard focus focus focus for my advisor all the time yeah it's a very common all the time yeah it's a very common thing yeah it is it is because that's thing yeah it is it is because that's it's it's a common thing for PCs vision it's it's a common thing for PCs vision this what's the vision right like you this what's the vision right like you need to focus yeah i mean maybe you to need to focus yeah i mean maybe you to focus your robots right exactly so i was focus your robots right exactly so i was interested in interested in in a whole range of things and which was in a whole range of things and which was also why already during my second third also why already during my second third year of PhD I thought some of this stuff year of PhD I thought some of this stuff that was I was working on my colleagues that was I was working on my colleagues were working on could be commercialized were working on could be commercialized mm-hmm so I i went ahead and this was mm-hmm so I i went ahead and this was back in two thousand eight and started a back in two thousand eight and started a computer vision startup based out of computer vision startup based out of stock of where i was doing my PhD and stock of where i was doing my PhD and they start up called oculus AI well the they start up called oculus AI well the street market and peyton to this yeah street market and peyton to this yeah yeah i wish we had yeah yeah there's yeah i wish we had yeah yeah there's another Oh Calista three everyone else another Oh Calista three everyone else about that wasn't us all right um but about that wasn't us all right um but but we we were active for five years we but we we were active for five years we worked on three different products or worked on three different products or rather I should say we pivoted three rather I should say we pivoted three times during that journey because we or times during that journey because we or two times because we realized were two times because we realized were first-time entrepreneurs we realized first-time entrepreneurs we realized that a lot of things we were assuming in that a lot of things we were assuming in the beginning we're not right but we the beginning we're not right but we worked in that start up on everything worked in that start up on everything from image moderation to logotype from image moderation to logotype detection in images to mobile product detection in images to mobile product recognition we built action app that I recognition we built action app that I used to call still call Shazam for for used to call still call Shazam for for products all this yeah i work for shoes products all this yeah i work for shoes bags and hand wrist watches oh nice yeah bags and hand wrist watches oh nice yeah and it was in towards the end of that and it was in towards the end of that journey I mean we were the startup was journey I mean we were the startup was by that time 15 people and we were still by that time 15 people and we were still not profitable and the investors and not profitable and the investors and VC's were kind of getting tired of that VC's were kind of getting tired of that and around that same time meltwater came and around that same time meltwater came into the picture as I said we were into the picture as I said we were already doing logotype recognition already doing logotype recognition detection in the wild meltwater being a detection in the wild meltwater being a media monitoring company was interested media monitoring company was interested in that in addition to that particular in that in addition to that particular technology they were also even more i technology they were also even more i would say interested in the team that we would say interested in the team that we had we had a team a strong team that I'd had we had a team a strong team that I'd built as founder there of 10 research built as founder there of 10 research engineers and computer scientists that engineers and computer scientists that had done PhDs in various fields of had done PhDs in various fields of computer science like computer science like so it was a very strong and well fused so it was a very strong and well fused team that meltwater kind of wanted to team that meltwater kind of wanted to copy paste into their own organization copy paste into their own organization to work with a whole range of data to work with a whole range of data science related machine learning related science related machine learning related problems and that was back in 2012 13 13 problems and that was back in 2012 13 13 was what was when the deal was finalized was what was when the deal was finalized and and me and the team were transferred and and me and the team were transferred to not water nice nice and so I know you to not water nice nice and so I know you operate basically distributed team yes operate basically distributed team yes or you are in Stockholm and London or you are in Stockholm and London you're in san francisco right so the you're in san francisco right so the engineering team is pretty distributed engineering team is pretty distributed all across the organization my all across the organization my particular T in the data science NOP particular T in the data science NOP division is in distributed across division is in distributed across Stockholm Budapest bangalore and here in Stockholm Budapest bangalore and here in san francisco interesting so yeah I I san francisco interesting so yeah I I managed 23 people that are quite spread managed 23 people that are quite spread out how do you make this like a he out how do you make this like a he seemed and like how do you structure seemed and like how do you structure problems soaked and if you can move in problems soaked and if you can move in peril and all this different place right peril and all this different place right yeah I I strongly believe ml water yeah I I strongly believe ml water strongly believes in colocation I mean strongly believes in colocation I mean we have 60 offices in about 30 countries we have 60 offices in about 30 countries which is more officer than Oracle has which is more officer than Oracle has and the reason is not that we're bigger and the reason is not that we're bigger than or we're not we're about thousand than or we're not we're about thousand people mm-hmm so 60 offices of thousand people mm-hmm so 60 offices of thousand people is it might seem like a lot but people is it might seem like a lot but that's because we really believe in that's because we really believe in colocation so but colocation the sense colocation so but colocation the sense of agile lean small teams right it of agile lean small teams right it colocation doesn't mean anything if you colocation doesn't mean anything if you have thousand people in one building have thousand people in one building right there just you know physically right there just you know physically present but they don't really work present but they don't really work together that's right so we like to have together that's right so we like to have small offices and small teams and in my small offices and small teams and in my organization in the data science organization in the data science department what we have done is that we department what we have done is that we have divided the work across these four have divided the work across these four offices in four different teams really offices in four different teams really and scrum teams agile teams that we have and scrum teams agile teams that we have and each of those have a very and each of those have a very well-defined task within the product well-defined task within the product stack mm-hmm that we work on and that stack mm-hmm that we work on and that way they can focus on that while being way they can focus on that while being part of the of course the the big part of the of course the the big picture okay so it sources directly this picture okay so it sources directly this science right how do you science right how do you choose the balance of forward-looking choose the balance of forward-looking R&D and kind of improvements in R&D and kind of improvements in operational improvements customer operational improvements customer requests how do you pick what is the requests how do you pick what is the next area of research you know what will next area of research you know what will help you kind of a blue the whole help you kind of a blue the whole business right really the the way I I business right really the the way I I think in any organization the way that think in any organization the way that needs to be designed is adapted to the needs to be designed is adapted to the ecosystem and the situation you're ecosystem and the situation you're you're in as a business we met water you're in as a business we met water have decided for now we want to dedicate have decided for now we want to dedicate about twenty percent of our efforts to about twenty percent of our efforts to forward-looking potentially throw away a forward-looking potentially throw away a research right research that may or may research right research that may or may not you know yield something but but not you know yield something but but still something we want to invest in and still something we want to invest in and kind of venture I think the trade-off kind of venture I think the trade-off between what I always have called the between what I always have called the exploration versus exploitation is exploration versus exploitation is something that companies especially rd something that companies especially rd every companies need to take seriously every companies need to take seriously it's a trade-off that you know either it's a trade-off that you know either way that you decide to balance that you way that you decide to balance that you that has to be very active and and and that has to be very active and and and thought well thought through trade-off thought well thought through trade-off imbalance that you do ours is in rough imbalance that you do ours is in rough numbers 2080 and and and what we decide numbers 2080 and and and what we decide that that twenty percent is it supposed that that twenty percent is it supposed to be is actually a function of to be is actually a function of bottom-up innovation within the bottom-up innovation within the organization that we don't want to organization that we don't want to hinder because we strongly believe that hinder because we strongly believe that you know if we're going to be you know if we're going to be innovation-driven we have to allow for innovation-driven we have to allow for bottom-up innovation and as some of it bottom-up innovation and as some of it is of course directed or at least hinted is of course directed or at least hinted within certain light constraints of what within certain light constraints of what we believe the market at a strategic we believe the market at a strategic level with product management believe level with product management believe that the market is is moving towards that the market is is moving towards within the next three to five this one within the next three to five this one and all work for instance on the and all work for instance on the knowledge knowledge based knowledge knowledge knowledge based knowledge graph that we were building this is one graph that we were building this is one of those efforts interesting yeah we of those efforts interesting yeah we talked about it before so this version talked about it before so this version so you're 2080 pizza is different take so you're 2080 pizza is different take on Google a twenty percent i write this on Google a twenty percent i write this kind of you kind of switched to a kind kind of you kind of switched to a kind of innovation and of innovation and in structure in such a way rest are you in structure in such a way rest are you talking to that interesting so yeah also talking to that interesting so yeah also another is very interesting right so another is very interesting right so this it looks like a lot of discipline this it looks like a lot of discipline or machine learning data mining AI in or machine learning data mining AI in many areas convert you know the fact many areas convert you know the fact that we need kind of or knowledge right that we need kind of or knowledge right so Google has its own system right and so Google has its own system right and kind of you know and everybody is trying kind of you know and everybody is trying to build this so so what do you think to build this so so what do you think across most interesting for you in this across most interesting for you in this in this next stage like you know what do in this next stage like you know what do you think like what excites you the most you think like what excites you the most particular building is knowledge as it particular building is knowledge as it applies to melt water right right i applies to melt water right right i think the the knowledge graph or as I think the the knowledge graph or as I sometimes refer to the business graph sometimes refer to the business graph because that's sort of the context because that's sort of the context that's relevant from that water where that's relevant from that water where the entities in the graph are actual the entities in the graph are actual business entities products people business entities products people organizations it's it's going to be a organizations it's it's going to be a total game changer it's it's going to total game changer it's it's going to not only redefine the landscape of not only redefine the landscape of business intelligence as we know it but business intelligence as we know it but it's also also going to redefine the the it's also also going to redefine the the landscape of search I mean if you look landscape of search I mean if you look at as you said companies that are not in at as you said companies that are not in business intelligence in particular like business intelligence in particular like like Google hmm or even Facebook right like Google hmm or even Facebook right they are taking the graph approach the they are taking the graph approach the knowledge graph approach very seriously knowledge graph approach very seriously I mean Facebook has that built in into I mean Facebook has that built in into their very existence and infrastructure their very existence and infrastructure and business logic so so I think having and business logic so so I think having having the big and small actors on the having the big and small actors on the web to move towards a graph way of web to move towards a graph way of thinking where you have your core thinking where you have your core entities in our case its business entities in our case its business entities but any in facebook cases entities but any in facebook cases people it's you user profiles York or people it's you user profiles York or entities being connected in a graph way entities being connected in a graph way with attributes and relationships is the with attributes and relationships is the way we're going to create a semantic web way we're going to create a semantic web right now that's the way then right now that's the way then unstructured you know plethora of unstructured you know plethora of information that it's totally exploding information that it's totally exploding around us is going to eventually find around us is going to eventually find the grid or the mesh in which it's all the grid or the mesh in which it's all advice into actual business value advice into actual business value and and at not water we we see this as a and and at not water we we see this as a long-term vision we don't expect to long-term vision we don't expect to solve this within the next couple of solve this within the next couple of years instead we want to be one of the years instead we want to be one of the players that lays some of the foundation players that lays some of the foundation blocks in this in this journey cool no blocks in this in this journey cool no this is great then it would be great to this is great then it would be great to collaborate I think you know the result collaborate I think you know the result of contributions of a source Crimmins of contributions of a source Crimmins you can do to this and crowdsourcing you can do to this and crowdsourcing obviously is one of their like most obviously is one of their like most efficient ways to create this kind of efficient ways to create this kind of graph so so really great to follow up graph so so really great to follow up and you know if anybody watching this and you know if anybody watching this you know fall on our meetup wants to you know fall on our meetup wants to help you know contact us as meetup help you know contact us as meetup organizers you know Catholic buy back organizers you know Catholic buy back and can hopefully we'll figure out some and can hopefully we'll figure out some some interesting ways because you know some interesting ways because you know with nitra you know we have documents with nitra you know we have documents but we also face the same issue we have but we also face the same issue we have names companies businesses right so a names companies businesses right so a lot of players basically and you know lot of players basically and you know women industries right so why women industries right so why not competing directly right so I think not competing directly right so I think for many companies make a lot of sense I for many companies make a lot of sense I choose to Columbia here so I would rip choose to Columbia here so I would rip up with kind of kind of philosophical up with kind of kind of philosophical questions so so you you I was like questions so so you you I was like you're a startup guy you have your own you're a startup guy you have your own company or you know your scientists your company or you know your scientists your entrepreneur and then you joined not entrepreneur and then you joined not water which is this big global water which is this big global corporation and you've been here for corporation and you've been here for several years so what is it you know several years so what is it you know which attracted to you what is exciting which attracted to you what is exciting for you in this setup would does not for you in this setup would does not water enable you to achieve because it water enable you to achieve because it amplifies you have a team then you have amplifies you have a team then you have you know really interesting business you know really interesting business like what is the most satisfying like what is the most satisfying interesting thing for you which are able interesting thing for you which are able to do in a setting like this it's of to do in a setting like this it's of course the sort of the fundamental course the sort of the fundamental question that any entrepreneur faces question that any entrepreneur faces between you know big corporate and kind between you know big corporate and kind of small small startups and then both of small small startups and then both have of course their their their upsides have of course their their their upsides I think for me and I've been at my water I think for me and I've been at my water for two years now what what i've for two years now what what i've discovered is that there is a side to discovered is that there is a side to the big corporate machinery that's the big corporate machinery that's really attractive for really attractive for an entrepreneur a startup person as well an entrepreneur a startup person as well especially at mal water not what I especially at mal water not what I really we like to see ourselves as a as really we like to see ourselves as a as the world's largest startup mm-hmm and the world's largest startup mm-hmm and and I think it's it's probably not and I think it's it's probably not unique to melt water you so what one of unique to melt water you so what one of the oldest ones exactly exactly it a lot the oldest ones exactly exactly it a lot of a lot of big companies out there are of a lot of big companies out there are embracing I mean I we've all seen what embracing I mean I we've all seen what iBM is doing with Watson the motion iBM is doing with Watson the motion technology a lot of big companies big technology a lot of big companies big corporations of the past are embracing corporations of the past are embracing the startup thinking and would start up the startup thinking and would start up thinking here I really mean you know thinking here I really mean you know enabling innovation from the ground up enabling innovation from the ground up in a way where short-term profits are in a way where short-term profits are not necessarily your first priority or not necessarily your first priority or even your second priority it's more you even your second priority it's more you know enabling things that will later be know enabling things that will later be creating profits mhm perhaps and perhaps creating profits mhm perhaps and perhaps not but you are willing to take that not but you are willing to take that risk and I think Mel water is a company risk and I think Mel water is a company that taste those risks and and takes that taste those risks and and takes them you know calculated and of course them you know calculated and of course what a big company like my water offers what a big company like my water offers you is resources in a way that you of you is resources in a way that you of course are the starter don't always have course are the starter don't always have right you have there's always the right you have there's always the solution of throwing money at something solution of throwing money at something which which can be very attractive which which can be very attractive specially if you have components if specially if you have components if you're broken down your problem into you you're broken down your problem into you know sub problems as good problem know sub problems as good problem solvers do some of the sub problems solvers do some of the sub problems might be less interesting to solve you might be less interesting to solve you want them to um you almost want to want them to um you almost want to assume them solved so you can focus on assume them solved so you can focus on the meat meat right then the meaty part the meat meat right then the meaty part of your problem if you can solve those of your problem if you can solve those subproblems by throwing money at them subproblems by throwing money at them say you can do you know you desperately say you can do you know you desperately now need named netted wreckage in now need named netted wreckage in Chinese in order to do something else Chinese in order to do something else with that you can you can solve that with that you can you can solve that with with a vendor but you have the with with a vendor but you have the money and met water we we like to focus money and met water we we like to focus on the problems that are interesting for on the problems that are interesting for us and our customers mm-hmm and us and our customers mm-hmm and therefore always have the option of therefore always have the option of solving something by just having having solving something by just having having a bag of money that will be just for a bag of money that will be just for dedicated sub problem which special is a dedicated sub problem which special is a trade-off between trade-off between right it is afraid of and you can focus right it is afraid of and you can focus your best vines on the stuff which is your best vines on the stuff which is really interesting exactly this is great really interesting exactly this is great yeah I noticed this you know also kind yeah I noticed this you know also kind of you know b2b companies I think I'm of you know b2b companies I think I'm most efficient and supporting R&D and most efficient and supporting R&D and like you mentioned you know it may not like you mentioned you know it may not work out where a start people die but work out where a start people die but you know a company will just you know you know a company will just you know switch to the next project right so soon switch to the next project right so soon in the wake and if it's the best setup in the wake and if it's the best setup when we have a start-up within a company when we have a start-up within a company right so which seems you guys like you right so which seems you guys like you guys have so that's also that's very guys have so that's also that's very exciting so you know thank you very much exciting so you know thank you very much for looking forward to your talk well so for looking forward to your talk well so my hands hurt us thank you sounds power