Meghma Chakraborty

Used streamlit to create a web app for automating database queries

•Built a GBDT-based churn model (78.7% macro F1, 3.16% MAE) with roll-forward cross-validation to prevent data leakage, using SHAP for interpretability and uplift trees for interaction analysis. Calibrated outputs (Isotonic Regression), optimized thresholds, and evaluated class balancing techniques.

•Worked on advanced text matching solutions using LLMs deployed via Hendrix Compute, enhancing large-scale entity resolution workflows.

•In parallel, generated data-driven ad business insights, built pipelines using Python and Big Query Runner, and developed end-to-end dashboards in Tableau and Looker. Leveraged Spotify brand-lift studies, handled ad-hoc analytics, and managed full-cycle data retrieval from Google Cloud for reporting and analysis.

• Developed a forecasting engine and optimized models (Prophet, ARIMA) and laid groundwork for deep learning models using PyTorch for predictive analytics.
• Engineered scalable ETL pipelines and automated metadata workflows using Python, VBA, and SQL across multi-service platforms.
• Led analytics initiatives within Contract Data Management (CDM), applying statistical modeling, revenue forecasting, and variance detection.
• Built Power BI dashboards and Power Apps for real-time visualization and stakeholder decision support.
• Contributed to NLP-driven contract parsing prototypes and explored neural network architectures for forecasting and anomaly detection.
• Advocated for MLOps best practices and collaborated on scalable data infrastructure to support future deep learning deployments.

• Product Categorization on spend data using LLM model and providing confidence by bootstrapping- Stood 6th in entire intelligence Operations.

• Developed credit risk models using anonymized European loan data to predict Loss Given Default (LGD) and Probability of Default (PD) using XGBoost; achieved 83% accuracy and 89% ROC AUC for PD prediction, and R² of 0.87 for LGD estimation, enabling data-driven risk assessment.

• Used Dataset from Yahoo Finance, predicted close prices.
• Visualizations of the data, Used ARIMA model for estimation and forecasting.
• Analyze volatility plots. Used ARCH & GARCH models for better fit of data.
• Further used GJR-GARCH model and plotted simulations for the next one year.
• Compared Model Performance using Information criteria values.

• Used data of Brent and WTI crude oil from FRED website & explored the data.
• Transformation of the data according to the model necessities.(ACF& PACF plots, ADF test)
• Performing both models(ARIMA and ANN), Compare between models, Forecasting using the best model.

• Worked on Capacity gap analyzer(RGA), to analyze the recruiter capacity gap by: Feature engineering, Time series modeling and forecasting using ARIMA, SARIMA, SARIMAX with Grid search CV for optimal model(using Python).
• Power BI- Made monthly data refresh and dashboard Formatting for Global data and North America customization.