Jul 2021 – Feb 2024 · 2 yrs 7 mos · Bengaluru, Karnataka, India · On-site

• Managing a team of Data Scientists, ML Engineers to i) release production-grade ML services for forecasting and anomaly detection, ii) contribute enhancements to Azure Machine Learning, Azure’s cloud Machine Learning service. Key achievements include:
• Design and validation of new ML algorithm for well-calibrated probabilistic time-series forecasting based on Deep Autoregressive Flows. The model was found to outperform Amazon’s DeepAR, Google’s TFT during offline testing. The REST service for training and inferencing based on underlying algorithm is being used by internal customers for demand forecasting,
• Working with Azure Machine Learning to enhance their Automated ML offerings. These include: i) design of AutoML service for multimodal training ii) enhance AutoML for NLP to process long sequence text input during training and inferencing
• Recent Publications:
• “Go with the Flow”: High Fidelity Probabilistic Time-Series Forecasting with TransFlow, a Deep Autoregressive Flow model (Microsoft Journal of Applied Research, Vol 17)
• Ultron-AutoML: an open-source, distributed, scalable framework for efficient hyper-parameter optimization (IEEE Big Data’20)
• Hyper-parameter optimization with REINFORCE and Transformers (IEEE Big Data’20)
• Recent Patents
• Techniques for generating a model for time-series forecasting (Microsoft patent filed with Patent Office on 5/27/22)
• Method and System to extract data dependencies for ML models (Microsoft patent filed with Patent Office on 5/6/22)
• System for tuning ML model hyper-parameters (Walmart patent filed with Patent office in 2020
• System for scalable ML model training and deployment (Walmart patent filed with Patent office in 2020)

Nov 2018 – Jun 2021 · 2 yrs 7 mos · Bengaluru Area, India

• Lead a team of 10+ Data Scientists/ML Engineers to design and deploy AI solutions across e-commerce verticals such as e-commerce catalogue, display ad targeting. Key achievements include:
• Ultron: A Kubernetes based distributed ML platform that enables a) hyper-parameter tuning, training & deployment of ML models at scale
• The platform has been leveraged by Data Science groups across Walmart to train & deploy over 500 ML models in production.
• The platform has halved model development cycle time, increased accuracy of trained ML models by 5%, and slashed compute cost of training ML models by 70%.
• Designed new hyper-parameter optimization algorithm, ReMAADE (Reinforce and Masked Attention Autoregressive Density Estimators). This algorithm was found to be competitive with Bayesian Optimization and Random Search for identifying the best Text Classification ML models within a computational budget. The algorithm has been integrated with Ultron.
• Nominated for IIM-Bangalore’s i-LEAD leadership program for showing outstanding leadership

Oct 2015 – Oct 2018 · 3 yrs · India

• Leading a team of Data Scientists, Analytics Engineers, Product Managers, Big Data Engineers to design Data Science/Artificial Intelligence products and solutions for enterprise clients
• Data Science engagements for recent enterprise clients:
• Vistra Energy, AI algorithm design for optimal bidding strategies in power markets,
• Dallas, Texas (Oct ‘17 – March ‘18)
• Developed state-of-the art algorithms for optimal bidding strategies in competitive, auction based day-ahead energy markets. The algorithm uses Deep Reinforcement Learning techniques in a multi-agent competitive setting.
• Technology stack: Tensorflow
• Team Size: 2 Data Scientists, 3 Analytics Engineers
• CenturyLink, Design & Implementation of Marketing Analytics Platform
• Denver, Colorado (Oct '16 - March '17)
• Product Recommendation Engine: Designed a new algorithm, Augmented-ALS, for personalizing product recommendations and driving customer engagement. Worked with Big Data engineers to develop its Spark API for scalable model training, inference. The model has been deployed in production mode and drives product recommendations from 6000 SKUs for ~1.2 million customers.
• Multi-touch point Attribution (MTA) Model: Developed a graph based model to estimate contributions of various marketing channels towards product purchases. The model has been deployed for channel spend rationalization.
• Technology stack: Spark 1.1
• Team Size: 2 Data Scientists, 3 Analytics Engineers, 4 Big Data Engineers
• Internal Research Projects
• Directing efforts to develop an AI platform to support uses cases in Computer Vision (object detection, instance segmentation, image captioning) and NLP (opinion mining, document level classification)

Apr 2015 – Jul 2015 · 3 mos · New Delhi Area, India

• Raftaar.in is the world's largest Hindi content aggregator website, aggregating content from a variety of Hindi websites across the news, travel, healthcare and entertainment domains. I spearheaded efforts to strengthen the search and content recommendation back-end. Ongoing work includes development of a content recommender prototype based on collaborative filtering for reader specific content customization based on attributes (gender, age, location) and reader’s content consumption pattern (regression based latent factor models).
• Tool stack: ElasticSearch, Nutch, Hbase, Java SE

Oct 2014 – Sep 2015 · 11 mos · New Delhi Area, India

• Lead product design and development effort for web enabled location based analytics suite of tools with applications in retail planning
• Work included IT platform architecture design, deployment of machine learning algorithms for predictive modelling of location estimate accuracy, synthesis of output from GIS, image processing and econometrics experts, and implementation of APIs to expose the services to clients over the internet.
• The product prototype was demonstrated to top Industry executives at AC Nielsen’s product exhibition conclave.
• The tool stack used: PostGreSQL, PostGIS, Hibernate, Spring MVC for web application development, Google Maps for plotting and visualization, Javascript, jQuery

Apr 2011 – Dec 2016 · 5 yrs 8 mos · New Delhi Area, India

• Co-founded and managed India’s leading digital satire brand, www.theunrealtimes.com. The site publishes fake news reports, comic strips, mock transcripts, fake Facebook and Twitter conversations among other formats, marrying irreverent humour with biting satire in its commentary on current events.
• The site attracts over 5 lakh unique visitors per month, and has a huge social media footprint, comprising over 3 lakh Facebook followers and over 1.2 lakh followers for its Twitter handle, @TheUnRealTimes.
• Followers on Twitter include PM Narendra Modi, Finance Minister Arun Jaitley and 4 other Union Cabinet Ministers, leading academic and economists including Dr Bibek Debroy, journalists (Barkha Dutt, MJ Akbar, Shekhar Gupta, Swapan Dasgupta) and figures from the world of entertainment. The site has also nurtured a number of talented satirists who contribute regularly.
• Our work has been frequently referenced and shared by leading politicians and the mainstream media. Some of the work has also been syndicated to the India Today group.

Dec 2009 – Feb 2014 · 4 yrs 2 mos · New Delhi Area, India

• Set up shell company to provide policy research and analytics services with leading policy makers in the domain of governance and policy research. Projects completed:
• Client I, NewsFlicks, Smartphone/Mobile optimized News Platform, India Today Group (May ‘14 - Dec ‘14)
• Editorial and technical guidance for social media and search engine optimization
• Client II, Uday Singh, Member of Parliament, XIVth Lok Sabha (Oct ‘11 – Oct ‘12)
• Designed and implemented an extensive socio-economic household census covering over 3 lakh households in Purnia Lok Sabha constituency. Oversaw survey design, selection of machine readable survey forms, and training of over 500 surveyors, data ingestion and storage in RDBMS database and finally data analysis and presentation of insights.
• Designed and implemented a monitoring and evaluation system for implementation of key welfare schemes in Purnia
• Client III, Former Finance Minister Mr. Yashwant Sinha (Nov ‘10 – Feb ‘11)
• Worked with Former Finance Minister to prepare the Opposition’s shadow version of the Union Budget of India

May 2008 – Nov 2009 · 1 yr 6 mos · New Delhi Area, India

• Strategy Consultant, ITC e-choupal
• Designed business model for distributing health care products through ITC’s e-choupal network
• Conducted market segmentation for new FMCG product launch

Jan 2003 – Apr 2006 · 3 yrs 3 mos · New Jersey, USA

• Portfolio Analyst, Portfolio Planning Group
• Improved existing portfolio management analytics by introducing stochastic programming, Monte Calro simulation techniques
• Part of team that implemented Portfolio Planning IT System (being used across all J&J Pharmaceutical companies)
• Lead Decision Analyst, Decision Analysis & Portfolio Management (Aug. 05 – Jan. 06)
• Worked with drug development teams to develop risk-reward models