Sparsh Gupta Email & Phone Number

Mumbai, Maharashtra, India

Richardson, Texas, United States

•Rice Crop Prediction: Secured first place in the EY Open Science Data Challenge Phase 1 by predicting rice crop presence in 250 geo-locations in Vietnam using satellite data with 98% accuracy.•Food Brand Growth Analysis: Utilized advanced statistical analysis in SAS and Python on extensive scanner data to identify strategic opportunities for brand growth and risk mitigation through hypothesis testing and regression techniques.•Drug Response Prediction: Built an advanced ML system with tailored estimators, feature selection, ensemble techniques, and SHAP values for superior interpretability, employing XGBoost, SHAP, and W&B.• RAG System: Developed and evaluated Retrieval Augmented Generation pipelines with semantic search, vector databases, and experiment tracking, using Langchain, Redis, Ragas, and OpenAI LLMs integrated with Weights and Biases.• Topic Modeling of Religious Scriptures: Engineered a clustering system utilizing BertTopic, HuggingFace, Pandas, and Spacy to categorize and visualize diverse philosophies within Hinduism.• Customer Analytics: Employed statistical models, Bayesian inference, and Zero-Inflated Count models in R to analyze customer behavior, churn probabilities, customer survival rates, and forecast site visits from customer characteristic data. Utilized SQL for relational data modeling and query writing focused on customer/order tracking and analytics.• Cricket Player Performance Prediction: Leveraged LightGBM and XGBoost with MongoDB data pre-processing to achieve an 85% increase in player performance prediction accuracy and 58% better strategic decisions during MC simulations.

Bangalore Urban, Karnataka, India

• Optimized model training and inference speed with multi-GPU utilization for online learning in production through micro-services.• Delivered 97% accuracy in market sentiment analysis by optimizing SOTA transformers (BERT, BART, XLNet) with HuggingFace and Spacy for Twitter, news, and report sentiment analysis.• Achieved a 98% MAPE in cross-validation and 96.5% in production for commodity price forecasting using a PyTorch Time-Series Transformer model.• Attained a production F1-score of 0.74 by leveraging deep learning techniques in Keras with TensorFlow and W&B to accurately detect anomalous temperature rises in smelters.• Boosted performance by 48% and faster inference by 122% in smelter temperature anomaly detection via PyTorch with LSTMs.• Saved $600K+ by empowering decision-making for traders and risk managers through the development of Python-based quantitative models on AWS S3 for commodity derivatives pricing, hedging, and risk assessment.

India

• Delivered technical insights and business strategies to internal teams while effectively articulating model performance, troubleshooting, and improvement ideas to clients. Enhanced model training pipelines on AWS EC2 with parallelization, resulting in a 61% increase in experimentation efficiency.• Backed trading decisions in Freight Forward Agreements and saved $250K through a HumbleAI ensemble learning system in Python.• Architected a top-ranked autonomous racing car in the AWS DeepRacer 2019 competition, leveragingProximal Policy Optimization RL model on AWS infrastructure.

Master Of Science - Ms, Business Analytics, 3.75

Coursework Predictive Analytics | Modeling for Business Analytics (Bayesian Modeling for Customer Analytics) | Advanced Statistics |.

Bachelor Of Engineering - Be, Chemical Engineering

Activities and Societies: Ultimate Frisbee Club Course work: Calculus-I,II | Differential Equations | Computer Programming (C/C++) |.