📉 Customer Churn Predictor
Classification
LogReg / Random Forest
EDA
Predict which customers are likely to leave and why. Clear pipeline, solid baseline scores,
and insights like tenure, contract type, and billing patterns driving churn.
Highlights
• Binary encoding + OHE · Stratified split
• ROC-AUC ~0.84 · Accuracy ~80%
• Feature importance for business actions
🚢 Titanic Survival Model
Logistic Regression
Kaggle
EDA
Predict survival using a clean, reusable baseline. Full write-up, styled tables, and a
notebook preview. Public leaderboard score at 0.76.
Highlights
• Missing value handling and encoding
• Validation accuracy ~81%
• Kaggle submission and results
📰 Fake News Detector (Deployed)
NLP
TF-IDF
LinearSVC
Flask + Gunicorn
A style-based classifier trained on fake vs real news using TF-IDF word/char features and LinearSVC.
Productionized on a VPS with a clean UI and solid evaluation.
What I built
• Pipeline: preprocessing → TF-IDF → baselines → evaluation → Flask API → deploy
• Robust vectorization, confusion matrix, artifacts
• Public demo with reverse proxy
🧾 CV Generator (Deployed)
Flask
FPDF
VPS
A fast, simple web app that turns form inputs into a polished PDF CV. Built with Flask + FPDF and deployed
with Gunicorn behind Apache on a VPS.
What I built
• Clean form validation and PDF templates
• Reverse-proxy setup and process management
• Zero-JS fallback for compatibility
📈 Time Series Forecasting
LSTM
Prophet
Backtesting
Prototyping deep learning and statistical approaches for financial series with proper backtesting,
leakage control, and interpretable diagnostics.
What I built
• Sliding-window training · walk-forward validation
• Error analysis and residual diagnostics
• Comparisons across ARIMA/Prophet/LSTM
🧠 AI Portfolio Website
Product
UI
Deploy
Hub for interactive demos, writeups, and model cards. Built for fast loads and clean design.
Roadmap
• Add blog posts with code snippets
• Ship a small MLOps tutorial section
• More interactive NLP demos