Sep 2019 – Dec 2022 · 3 yrs 3 mos · Bengaluru Area, India

• AUTOML AND AUTOEDA FOR DEMAND FORECASTING
• Created end-to-end Auto-EDA and Auto-ML python packages for demand prediction and deployed the tool using docker images
• Applied data imputation and transformation though pre-defined input schema formats for automatic feature engineering
• Generated impactful analytics presentations leveraging insights using matlotlib, pandas-profiling along with Hypothesis testing
• Ran multiple open source regression techniques like linear regression, tree based regressors, XGboost and Neural Nets; Statistical
• models like HoltWinters, Lew, MA and in-house Cyclic Booting using multiprocessing and hyper parameter tuning
• Compared multiple accuracy metrics and created post model analysis plots to get best modelling results
• LUMINATE DEMAND EDGE
• Operated end-to-end client project with in-grown Cyclic Boosting algorithm to replace time series modelling in Dask and Python
• Created customer specific features like promotional events, weather at geo-location, covid and achieved 20% improved accuracy
• Actively involved in migration of the product to Azure environment and deployment using docker and mlflow
• MARKDOWN PRICING OPTIMIZATION AND DEMAND FORECASTING
• Worked on optimization model of price for different customer based on custom strategies for customers
• Created covid modelling features by adding regularization for covid period for demand forecasting
• Worked on Demand forecasting with cyclic boosting algorithm and compared it with Lewndowski, Holt Winters and other statistical and ML models.
• Worked on deployment of model in production with Airflow, FastApi and model versioning with MLflow.

Aug 2018 – Sep 2019 · 1 yr 1 mo · Bengaluru, Karnataka, India

• DIGITAL IDENTITY AND RISK (Credit Card Application Fraud Detection)
• Engineered features based on device geo-spatial data and credit card application data for US market in Apache Spark
• Handled class imbalance in model by using advanced oversampling techniques like ROS, EasyEnsemble along with SMOTE
• Tuned classification models using grid search to get LightGBM as best performing model to predict fraud case for application
• GEO-FENCING BUSINESS BOUNDARY POLYGONS AND ENTITY RESOLUTION
• Applied image processing techniques to identify the edges of building in Maps image and identified other businesses using Optical Character Recognition
• Created labelled dataset for identifying if a business is still operating and applied XGBoost model for classification

Jul 2016 – Aug 2018 · 2 yrs 1 mo · Gurgaon, India

• NETWORK SIMULATOR
• Estimated probability of shipment to breach TAT using GB Regression and updated these probabilities at each scan
• Fitted statistical distributions of error over time taken by shipments in long haul movements in Apache Spark
• LOCALITY BOUNDARY POLYGONS
• Replicated geographical city locality boundaries on maps using transportation GPS data and Open Source Routing Machine
• Built geofencing system for Address Correction Engine and applied linear regression model to predict time spent to deliver shipment in each locality using different features

May 2014 – Jul 2014 · 2 mos · Rostock Area, Germany

• Analysed Xu’s Bayesian regression method for estimating polygenic effects in Fortran language
• Implemented sparse storage scheme for inverse correlation matrix to estimate trait differences by pairs of markers
• Achieved improvement in trait detection results by 10% using chi square distribution to estimate variances

Dec 2013 – Dec 2013 · 0 mo · Mumbai Area, India

• Analysed international models for Decentralization of Power, Referendum with inclusion in National Manifesto of AAP
• Analysed case studies of world class cities to understand most desirable pattern of urban planning in Mumbai
• Carried out innovative political campaigns on streets of Mumbai and adjoining villages