📈MACHINE LEARNING: FROM ENGINEERED FEATURES TO FORECASTS
📈MACHINE LEARNING: FROM ENGINEERED FEATURES TO FORECASTS
Embedded Files
Built predictive models that explain variance, surface key drivers, and deliver interpretable insights.
Built predictive models that explain variance, surface key drivers, and deliver interpretable insights.
SavTankVision — AI Classifier for Co.'s Photo Archive
SavTankVision — AI Classifier for Co.'s Photo Archive
AI pipeline that finds shipping photos in job folders, uses CNN to determine if image is a detail image or a full tank image, and centralizes them for fast reuse.
Result: Processed ~15,000 photos across 823 jobs (2019–2025). Staff can pull reference images in minutes instead of hours; originals preserved; re-run safe.
Tech used: Python, PyTorch (ResNet18), torchvision, Pillow, pandas, pathlib, scikit-learn (kNN baseline), OpenCV (early preprocessing).
My role: Data scientist / workflow engineering — built crawler, trained CNN (upgraded from kNN), added copy-only pipeline with logging and duplicate avoidance.
🏘️ Real Estate Price Prediction
🏘️ Real Estate Price Prediction
Result: GBT explained 71% of the price prediction variance across 2M+ home listings
Result: GBT explained 71% of the price prediction variance across 2M+ home listings
Tech used: Python, PySpark (MLlib), pandas, matplotlib, seaborn, SciPy, Google Colab
📦 Includes: VectorAssembler, GBTRegressor, RegressionEvaluator, and statistical testing (ks_2samp)
🏎️ What’s a Constructor?
In Formula 1, a constructor earns points based on both of its drivers' results. This analysis uncovered which metrics — like grid position and lap consistency — best explain overall team performance.
🏎️ F1 Constructor KPIs
🏎️ F1 Constructor KPIs
What it solves: Identifies which factors matter most in determining a team’s success in Formula 1. Goes beyond lap times to uncover constructor strategy drivers.
Tech used: R, caret, ensemble models, tidyverse, constructor-level modeling
UX optimization using statistical testing
UX optimization using statistical testing
🎯 Customer Service A/B Test
🎯 Customer Service A/B Test
What it solves: Designed to demonstrate how a UX design change impacts customer behavior, A/B testing logic and statistical rigor.
What it solves: Designed to demonstrate how a UX design change impacts customer behavior, A/B testing logic and statistical rigor.
What I built: A clean, replicable A/B test using Python. Included proper test group separation, t-tests, and recommendation logic based on statistically significant metrics.
Tech used: Python, pandas, scipy, statistical testing
🟡GitHub
Page updated
Google Sites
Report abuse