Junior Data Scientist and Statistician

Authored the Volume Programming Methodology for petroleum products and LPG, a 73-page statistical workflow proposed to replace the previous 2017 procedure. Designed a traceable and reproducible nine-stage methodology based on SARIMA/SARIMAX models, intervention analysis, Box-Cox transformation, ADF and KPSS testing, AICc selection, residual diagnostics, interval forecasting, and anomaly detection.

Built the operational Python solution using pandas and statsmodels, including an analyst notebook, 10 reusable modules, and automated Word report generation. Produced theoretical documentation, installation and operation manuals, interpretation guidance, and an Excel template with data-quality controls.

Led a four-person team developing a calibrated XGBoost classifier to predict next-day rainfall from humidity and pressure data. Created 19 leakage-free engineered features using 378 days of weather-station observations and applied Platt scaling for probability calibration.

Audited the preprocessing pipeline after an external review and corrected critical data issues, including duplicated records, sensor interpolation, rainfall-intensity aggregation, and temporal leakage. Evaluated the model with date-based holdout, TimeSeriesSplit, and a 295-day walk-forward evaluation against four baselines; documented that the model did not outperform the 30-day climatology baseline and advised against operational alert-system use. Delivered a reproducible package with 21 automated tests, 55 consistency checks, and a manual for non-technical farmers.

Final-semester Mathematics student, expected to graduate in 2027. Coursework includes probability, mathematical statistics, statistical inference, survey sampling in R, linear models, optimization, real and functional analysis, measure theory, algebra, numerical analysis, and databases. Authored lecture notes on inference with R, including probability sampling, Horvitz-Thompson estimation, and stratified sampling using official microdata.