Jul 2021 – Present · 4 yrs 10 mos · Hyderabad, Telangana, India · On-site

• Building and deploying scalable machine learning/GenAI based solutions in Microsoft's web experiece products like Bing, Edge, MSN etc.

Apr 2019 – May 2021 · 2 yrs 1 mo · Hyderabad Area, India

• Responsible for
• Building solutions to modernize workforce planning for Microsoft Consulting Services
• Developing models to drive growth in Azure Consumption Revenue
• Building automated frameworks for repetitive tasks to reduce the ML project life cycle
• Helping to build an org by hiring and mentoring new talent in the team
• Key Initiatives:
•  Modernizing Workforce Planning for consulting business (~10k resources)
• Enriching skill profile of consulting resources using collaborative filtering and association rules mining to get accurate estimate of available capacity.
• Leveraged the automated forecasting module (Custom built) to forecast demand for resources across various grains of Area, Domain, Role & Skills.
• Implemented a gap detection framework to identify gap between available supply and demand, and provide specific inputs to budgeting, planning, hiring & reskilling initiatives.
•  Built a recommendation engine to identify and staff best fit Consultants/Architects to the consulting delivery projects.
•  Built a Sequential Recommendation Algorithm to identify consulting services journey that can maximize Azure Revenue from a specific customer. This uses state transition concepts from Markov chains to derive causal patterns between consulting services and consumption growth.
•  Built a Scalable Automated Forecasting Suite with following features:
• Capability to clean & transform historical data and run ~20 base models & ensembles up to 5th order.
• Identifies best model based on error(MAPE), stability(Churn) and bias.
• Models are built in R/Python and Spark is used for scalability.
•  Built an automated Feature Engineering Tool for Classification Models with following features:
• Capability to clean input data, impute missing values, treat outliers, create engineered features.
• Also adds an easy-to-use UI to see the visuals of all the features and do custom adjustments.

Apr 2017 – Mar 2019 · 1 yr 11 mos · Mumbai, Maharashtra, India

• Responsible for-
• Statistical and Machine Learning Model Development(Acquisition/Collection/Geo Expansion) and Validation, for all the Lending Businesses(MicroLoans, TwoWheeler, Housing) as well as Wealth Management Business.
• Data-Driven Decisioning and Strategy Formulation for Micro Loans Business.
• Operational Cost Optimization by Deriving insights from Unstructured Text Data using text analytics techniques and overlaying them with quantitative data.
• Credit Loss Estimation for Optimum NPA Provisioning.
• Estimation of Cross-Sell and Upsell opportunities and development of new revenue channels.
• Managing Expectations of Multiple Stakeholders with timely delivery of projects in steep deadlines.
• Managing a team of 3 Data Scientists
• Impact: Numerous key decisions were taken based on the insights along with the implementation of Models, which contributed to doubling the Portfolio & Cutting delinquencies from 15% to 5% in 16 Months. 400 New Branches(Geo Expansion Project) were established contributing 30 % of total Business and a new line of Top-Up & Loyalty Products were Launched. I achieved Star-Award (one of the 16 people to receive) for my contribution to the growth story.
• Worked with Tools: Python, R, SQL, SAS, Tableau, MS Excel.
• Statistical and Machine Learning Techniques used:
• Supervised: Random Forest, Neural Network(ANN, RNN), GBM, SVM, XGBoost, Decision Tree, Naive Bayes, Logistic Regression, KNN, Bagging and Boosting Techniques
• Unsupervised: Clustering using Kmeans/Hierarchical/Density Based/Expectation Maximization, Collaborative Filtering
• Please take a look at the projects section for more detailed summaries.

Jul 2012 – Jul 2016 · 4 yrs · Odisha, India

• Responsible for-
• Setting up infrastructure for Digital & IOT Journey of the company.
• Electricity Demand Forecasting using ARIMA and Neural Networks for optimum Load scheduling and getting best deals in advance for distributing residue power to grid.
• Establishment of Energy Management System to optimize the use of energy. This comprised of energy data collection, cleaning, clustering into various groups based on different types of load, and formulation of cluster specific energy saving strategies.
• Design and implementation of Predictive Maintenance Techniques based on Triggers deduced from analysis of Historical Data to improve equipment life and reduce plant downtime.
• Development of a Set Points Master for PID Control Devices to get best operational efficiencies in terms of TAT and Output Quality.
• Impact-
• Energy consumption reduced remarkably, which helped JSL to receive Energy Excellence Award in Gold Category.
• Plant Downtime (due to electrical equipment failure) came down under 5% from 30 %.
• Very close demand forecasting helped us earn much more by making deals in advance for residue power, and there was huge reduction in penalties enforced by the Grid for not meeting the promised demand.