• Worked with Uber Graph and Risk ML teams to develop a generalized workflow for generating node embeddings using Relational Graph Convolutional Networks(RGCN) for heterogeneous graphs.• Used RGCN to predict a score for each payment instrument(PI) to determine its affinity towards being involved in a fraudulent transaction at trip-level.• Built the deployment pipeline to predict scores and embeddings for daily active payment instruments.

• Built a weakly-supervised NER system for automated extraction of medical vendors and device-modality pairs from published medical literature using WISER.• Integrated a feedback loop pipeline to regularly update the system based on the knowledge base of high confidence entities constructed using statistical methods.• Analyzed trends to capture the Mindshare of medical vendors and their devices.

• Worked as a part of Growth Platform team, aimed at targeting potential usersfor the MyGalaxy app using the current app users as the seed data.• Remodeled the look-alike modeling problem as a recommendation problemusing the Field-aware Factorization machine model with Implicit feedback.• Built the Graph Embeddings using FastText for user visited web domains to model userpreferences based on their web browsing history.• Predicted Adaptive Timeout for individual users… Show more • Worked as a part of Growth Platform team, aimed at targeting potential usersfor the MyGalaxy app using the current app users as the seed data.• Remodeled the look-alike modeling problem as a recommendation problemusing the Field-aware Factorization machine model with Implicit feedback.• Built the Graph Embeddings using FastText for user visited web domains to model userpreferences based on their web browsing history.• Predicted Adaptive Timeout for individual users using LSTM networks to boostAds Platform revenue while optimizing page speed to improve user experience. Show less

• Worked on constructing a knowledge base for an E-commerce Price Comparison Engine forSamsung Mall, which aimed at aggregating similar products from differente-commerce websites.• Extracted and normalized product attribute key-values pairs from unstructured product title anddescription using a linear-chain Conditional Random Fields sequence labelingmodel.• Utilized the data scraped from open source websites for enriching the structured attributes table.• Trained a… Show more • Worked on constructing a knowledge base for an E-commerce Price Comparison Engine forSamsung Mall, which aimed at aggregating similar products from differente-commerce websites.• Extracted and normalized product attribute key-values pairs from unstructured product title anddescription using a linear-chain Conditional Random Fields sequence labelingmodel.• Utilized the data scraped from open source websites for enriching the structured attributes table.• Trained a MaxEnt Classifier for the Matcher to classify a pair of product as similar, variants or different using the extracted structured attribute-value pairs as well as unstructured information for constructing the features vector.• Coordinated with the crawling data partner for the exchange of data.• Created a common category hierarchy for all the different e-commerce websites. Show less

• Compared the performances of different techniques of Collaborative Filtering(CF) recommendation systems for an App Recommendation Engine for SamsungGalaxy Apps.• Implemented the Item based and User Based CF recommendation systems usingPyspark from scratch and used Spark MLlib library for other CF approaches.• Obtained a maximum precision of 81.1% for User Based CF system using the TopN recommended items approach for evaluation.