Natural Gas

- Leveraged engineered features from geo-spatial, survey, and economic data to deploy 17 flagship mod- els estimating the supply of soft commodities. Led end-to-end processes encompassing coordination, planning, research, experimentation, deployment, monitoring and causal analysis.- Employed Docker containers in AWS Batch and ECS to execute experiments while leveraging a feature store, aiming to cut down AWS expenses and expedite grid search iterations. These endeavors led to the discovery of optimal hyperparameters for XGBoost, marked by reduced volatility and resulting in industry-leading model MAPEs ranging from 2-6%. Monitored models using Airflow and Databricks.- Orchestrated modeling team’s JIRA board, enhancing stakeholder coordination and project workflows.- Founding member of engineering hiring advocacy group leading to establishment of standardized recruitment processes, attracting 20+ outstanding engineers and data scientists.

- Built China's Winter wheat yield model based on data from NBS. (R^2: 0.93; MAPE: 2.27%)- Built China's Wheat yield model based on data from JCI. (R^2: 0.83; MAPE: 4.08%)- Built Canada’s Spring wheat yield model with in season prediction using geospatial and temporal features which cut the MAPE of the best Statcan model (8.6%) by more than half (4.0%).- Continue to monitor, debug and improve automated email alerts of Gro data-series.- Maintained code for Gro-derived and Gro-indices series estimating demand and supply of commodities.- On boarded new engineers by giving them Gro's engineering infrastructure overview.- Managed on-call duties and product deployment tasks as part of the manual deployment pipeline.

· Built an end-to-end API feature called Analogous Years to rank order time periods using time series analysis.

1. Consulted with Valor Water Analytics to find anomalous water meters (a $8.7B revenue loss problem).2. Used time-series forecasting, supervised classification, flask and AWS to solve and deliver a dashboard (http://waterprophet.xyz/).3. Valor will use my time series forecasting model for their water consumption prediction.

1. Recommendation system - used NLTK toolkit, Tfidf transformer, and a tokenizer package to preprocess and transform show metadata data into numerical data. Later, used LDA, LSI, NMF, KMeans algorithms to cluster transformed data. Finally, used Euclidean distance tocompute nearest neighbors of a given show.2. MIS Automation - Automated the daily reports for each markets that the company currently operates using a BI tool called Looker.3. ETL Process - Wrote SQL queries for parsing urls to extract paid search information to be loaded into the database.4. Predictive Modeling - Trained a predictive model to understand repeat purchase possibility of a customer based on their first purchase behavior.

1. Research in geometric topology, geometric group theory, formal language theory, natural language processing. 2. Proved a theorem that determines if there is hyperbolic behavior in the large scale geometry of a very general class of space of graphs.3. Taught more than 1,000 undergraduate students. Instructor to advanced undergraduate mathematics courses like linear algebra 350 (Linear Transformations, row reduction, eigenvalues, eigenvectors, diagonalization of matrices, applications to differential equations.); statistics 211 (binomial, Poisson, geometric, normal, chi-square and t-distributions. Confidence intervals and testing of hypothesis for mean, standard deviation and sample proportions.); Calculus-136 (Integration and applications, infinite series, polar co-ordinates.); Precalculus 114; Basic Calculus 119; College Algebra 109; College Algebra 107.