Vinay Vijay Sawant

Vinay Sawant  Current Position Software Engineer(Backend/Data Engineer/Data Science/ML) • City, MUMBAI IN •-I have around 7+ years of experience as Software Engineer(Java/Python/Golang) by working into multiple technologies in Ecommerce and Fintech Domain.Experience with data engineering, machine learning, deep learning. Love to work on building backend/data science,data engineering in its entirety GitHub : https://github.com/vinay10949 Contact No: -/ - Motive & Pro le Motive & Pro le Go-getter, passionate about technology, innovation. Eaager to utilize my technical skills combine with AI/ML domain knowledge and create practical products for the real world.Willing to develop values under company culture. Skills Programming Skills 7+ years of programming experience in languages like Java,Python,Golang. Experience on building BackEnd Systems,Microservices,Recommendation Engine, Loan Default Prediction Modules, SMS Pro ling ,Fashion Recommendation etc using Data and Machine Learning. Machine Learning skills Sklearn,Numpy,Pandas,rapids,scipy, Keras, Tensor ow ,and Big data analysis skill with Python on multiple scenarios. Have attended few Kaggle competition Technologies: Python ,SQL,MySQL ,Golang ,Java , Spring , Hibernate ,Machine Learning ,Deep Learning ,Hadoop , Hive, Keras , Scikit-learn, Tableau ,Linux ,Redis ,NoSQL ,NLP ,Docker ,Computer Vision ,Rest/Grpc API ,Apache AirFlow, Git Cloud Platform: Google Cloud Platform,AWS Work Experience SuperMoney (Software Engineer Backend,AI/ML/NLP ) June 2016 ~ Present Built backend powered with microservices using rest/grpc apis . Built complete backend. Built RealTime Noti cation Engine that powers customer reminder noti cations for repayments and eligibility for new loan product. Built Calibrated ML model that predicts the probability of customer likely to default a short term loan. Built Computer Vision Module for identify aadhar and pan card, cheque and doing OCR on to Pan and aadhar. Also built scalable distributed systems. Also built a simple SMS pro ling using regex. Software Engineer (Bookmyshow Aug 2014 ~ May 2016) Worked here as software engineer,worked in core team of building recommedation engine with low latency. Worked on building backend services. Worked on cold start problem of Music Recommendation Engine using audio features using vienna university library,then used machine learning to recommend music. Worked on bookmyshow realtime seat occupancy module that show real time seat occupancy on to bookmyshow app BI Consultant (Team Computers-Mercedes) Aug 2013 ~ Feb 2014 Worked for client Mercedes ,helped them with building marketing dashboards in Qlikview Data Science Projects Document Detection using CNN Developed a CNN model for identifying document types like PAN,AADHAR FRONT,AADHAR BACK,CHEQUE, Was responsible for model training ,creation and deployment. Yet to be productionized. CardioVascular Disease Detection(Hackathon May 2020) Github: https://github.com/vinay10949/CVD Role- End to End Deployment,Testing,Monitoring,Retraining Responsible Detailed EDA,Model Creation,deployment,testing of model,monitoring and retraining approach Achieved Recall of 82% ,using F2 score as metric. SmartBackGround Veri cation using Deep Learning,NLP and Computer Vision Identifying eye descriptors and then calculated EAR I.e eye aspect ratio for consecutive frame to identify if person in frame is blinking his eyes or not Built face veri cation module to verify pro le photos with document photos. Built name and address similarity between provided name and name in the document using combination of cosine similarity and Levenshtein Distance . Short Term Loan Default Prediction Objective was to perform data engineering ,data pipeline and detailed analysis of our supermoney customers. Also predict which rst time Uber Customer is likely to default 5000 Rupee product loan. Achievement : achieved 0.79 F-Beta Score , loan default rate dropped by 40%. Predict Mercedes-Benz Greener Manufacturing Con guration Testing time (Kaggle) Objective was to predict testing Time for various combination of congurations..This could be useful to priortize the congurations to be tested that could save most of our time. - Given a set of feature variables predict testing time. Github : https://github.com/vinay10949/AnalyticsAndML/tree/master/Kaggle/MercedesBenz_Greener_Manufacturing Quora Question Pair Similarity (Kaggle) Built a machine learning model to predict whether two questions asked on quora are similar or not . So that the similar questions asked may have the same answers which have been given earlier for the previously asked similar question. Github : https://github.com/vinay10949/AnalyticsAndML/tree/master/Kaggle/quora-question-pairs HomeDefaultCredit Risk(Kaggle 0.77 AUC) Challenge was given various data like applicants- information about each loan application at Home Credit) bureau,we had to predict what is the application's capability to repay the loan. Achievement : Was in top 30% Github : https://github.com/vinay10949/AnalyticsAndML/tree/master/Kaggle/HomeCreditRisk Fashion Out t Recommendation based on Content Recommendation 2019 Used a pretrained Vgg16 model to extract visual features from outt , extracted text features from descriptions of outt like like weighted tdfword2vec Used combination of above features to nd the immediate neighbours tried both using Euclidean distance,and also trained LSH model above high dimensional of combined features to nd immediate neighbours of out t. Futher future ongoing work: Collecting fashion documents to train a word2vec on all fashion corpus documents like vogue etc, Convert a user text or query into a elements,style document query. Use outt as sequence of items and train a sequential model as recurrent networks, or graph Neural networks to solve problem like recommending an outt as sequence of items,or if any one item is changed new sequence should be considered. Music Recommendation Engine(May 2016) Created ML Recommendation engine for music,used Music Information Retrieval (MIR) from libary by Vienna university,used features like rhythm pattern,temporal descriptor,etc(approv 2000dimensions of one song) Used Locality Senstive Hashing algorithm +Episilon Greedy Approach for recommendation Recommendation Engine based On History and Ratings (Dec 2014-July 2015) Developed Timeband rules for users booking their events on BookMyShow,Developed system which recommended nearby venues,Also developed Recommendation system which used Movie Ratings to generate recommedations,Calculated Bayesian average of reviews to recommend top events (Movie,Plays and Concerts) in a region Technology : Python,Redis,MongoDB,Php Achievement Built one core component for Recommendation Engine with very low latency Customer Segmentation depending on customers taste (Mar 2015) Segmented the customers depending on customers taste ,as in genres,languages using Unsupervised learning using Spectral Co Clustering on to m aggregated dataset of users of transaction,ratings data. .Validated the bicluster using consensus score Customer Segmentation based on transaction history(Feb 2015) Customer Segmentation based on transaction history (BigData) Feb-Mar 2015 Created a clusters of users based on their rules on transaction data, data was stored in MongoDB,user proles were over 10million docs,Users were classied as star user,active user,lapser,bouncer and OTB etc BackEnd Developer Projects Smart Video KYC (July2020 -Present) Created Golang gRPC powered micro-service for video kyc for Smart Background veri cation . Used protocal bu ers and created grpc based Unary API. Designed High Level Architecture, written docker script and CI /CD Pipeline. SuperMoney Entire Backend API (2016 -Present) Created and handling entire backend for product called SuperMoney,the entire rest apis uses Spring Hibernate. SMS Parsing(2016 ) Created SMS proling application using regular expression and Named Entity Recoginition using python NLTK Golang GRPC Microservices for Project Management System (Freelancing) 2018 Built robust grpc based microservices,built using Golanguage and GOA framework and developed corresponding test cases . Client ResearchNow. AscertLogger Microservices for AT&T(Freelancing) Built an highly scalable logger service using Golang that can handle million writes per 10sec,used channels extensively,Wrote code for log rotation ,log compression . Authentication services Freelancing July 2018 Ceated Golang grcpc webservice that uses vault to run authentication service.Wrote service that creates users,assign roles,create certicates,authenticate using certicate etc Client : F5 Authentication services Freelancing July 2017 Built Golang webservice that uses GoogleCloud pub sub for creating Notication system Wrote GRPC based webservice in golang and created processor which listens to subscriber and sends notication to users. Role- Was responsible for creating and deployment of system into docker end to end. Was also partly responsible for High Level System Design Client : Schlumberger GPS Web Service Built an logger service in Golang that can handle million writes of gps coordinates per 10sec,used channels extensively,Wrote code for detailed summary stats of gps coordinates like Distance travelled in a day , Which coordinate is likely to be his home. Role- Was responsible for creating and deployment of system into docker end to end. Was also partly responsible for High Level System Design Client : Supermoney SeatOccupancy Built a entire seat occupancy module that pinged 2500 cinemas by generated schedules,and capture the real time seat occupancy at that particular time ,all that information was stored in queuing system which was later consumed by SeatOccupancy engine that shows real time seat allocation in cinemas. Technology Nodejs,RabbitMQ,Mysql Client : Bookmyshow Education Masters in Computer Applications- Computer Science Major (Full Time) Bachelors in Computer Science- Computer Science Major Personal Details DOB :- Birth Place :Mumbai. Languages: English, Hindi and Marathi. Marital status : Married