DBTB INT Shourabh Rawat

yeah so I'm sorry bravas I work for trulia zillow group so I think data by the way is a nice conference in done this sort of covers a diverse set of topics and that allows me as speaker to sort of sort of present to a wider a wider range of audience and be getting to know like how usually like working on as a data scientist on a particular domain yourself do not see different perspectives that you people have two were the same problem on the same kind of data this speaking here allows you to sort of interact with those those people and sort of get their insides get their problems and get a better understanding so I think data is the reason why the school is because the data has there's a lot of data so first of all and to convert that data into information is is a major ingredient that is that is required to sort of make any business profitable or right since ever now nowadays like everything is going everything is online right to really get ahead of the game you need to understand your customers better your users better and each and every inch of their activity and so the day is sort of converting that data which is sort of rich in like they say the query logs or in terms of what they are interacting with is in is key key to sort of figuring out how you can multi monetize your business so the the main idea behind my talk was too like was to figure out like how to deploy image recognition systems at when you have a really small team so mostly especially in Atkins scenario where we have a data science team that wants to sort of provide image recognition as a service right and usually what happens is in standard environments usually have data scientists working separately from from the actual deployment architecture as a result what happens is that you will basically have batch processes and and iterations are much slower because then you have the data size have to explain their models to the to the deployment engineers to sort of sort of make that deployable with this job we said like image recognition is an ideal candidate where we can actually use an existing Python web stack and allow data scientist to seamlessly are easily sort of use that to allow for a sort of a near real-time sort of prediction engine and then provide that as a service to across across the company to sort of sort of conform to a like a service-oriented architecture where data science or image recognition becomes a service so Java talk was all around like how you can easily use like famous like celery and Django and build such deployment framework easily on top of your existing image ignition libraries so to become a good data scientist I think there should be we should have at the curiosity sort of two kno know about data so have to explore to explore new new articles new blogs and read about it understand it and being able to continuously experiment and try and sort of a trial and error and and sort of continuously build a build a better understanding of the data and create models that provide value value to us to your model to your company and to the users