Projects

Problem: A recipe site needed to predict which recipes would drive high traffic to optimize homepage content placement.

• Approach: Built classification models (Logistic Regression, Random Forest, XGBoost) on recipe attributes like category, servings, and nutritional data
• Result: Achieved 81% precision on high-traffic predictions, exceeding the 80% business target. Live interactive demo on Streamlit

Problem: E-commerce businesses lose revenue from customer churn, but lack early identification of at-risk users.

• Approach: End-to-end ML pipeline with EDA, feature engineering (RFM analysis, behavioral signals), and model comparison across Logistic Regression, Decision Tree, and Random Forest
• Result: Built a churn predictor that helps segment high-risk customers for targeted retention campaigns. Live demo on Streamlit

Problem: Raw Google Analytics exports contain missing values, inconsistent formats, and noisy data — making analysis unreliable.

• Approach: Built a full ETL pipeline handling null imputation, type casting, outlier detection, and automated visual reporting with Matplotlib and Seaborn
• Result: Transformed messy GA data into clean, analysis-ready datasets with 15+ automated visualizations. Deployed on Hugging Face Spaces

Problem: Marketing teams manually track campaign KPIs across spreadsheets, leading to delayed reporting and missed optimization opportunities.

• Approach: Built an interactive Streamlit dashboard with SQL-powered data aggregation, real-time filtering by campaign/channel, and automated ROI calculations
• Result: Dashboard consolidates key marketing metrics (CTR, CPA, ROAS) into a single view, reducing manual reporting time. Live demo available

Problem: Supermarket loyalty programs lack data-driven spending predictions to personalize customer offers effectively.

• Approach: DataCamp Associate Practical Exam project — applied advanced feature engineering on retail transaction data, compared multiple regression models
• Result: Delivered a predictive model for customer spending patterns with actionable insights for loyalty program optimization

Problem: Real estate teams rely on intuition for pricing, often leading to overpriced or undervalued listings.

• Approach: Explored 80+ features from residential property data, performed correlation analysis and feature selection, then trained Ridge and Lasso regression models
• Result: Built a price prediction model that helps agents set data-informed listing prices, reducing pricing guesswork

Problem: Data science learners often struggle to find a single, organized reference for the math behind ML algorithms.

• Approach: Curated and organized essential formulas across linear algebra, calculus, probability, and statistics with clear explanations and Python implementations
• Result: Open-source reference guide with 50+ formulas, deployed as a searchable static site on GitHub Pages

A developer portfolio and blog built with Next.js 16, Tailwind CSS v4, and shadcn/ui.

• Features: Dark mode, MDX blog with syntax highlighting, component registry, GitHub contribution graph, SEO optimization, and analytics
• Deployed on Vercel with perfect Lighthouse scores in performance and accessibility

Problem: Most meditation apps lack personalized, conversational guidance that adapts to user mood and preferences.

• Approach: Built a full-stack React + TypeScript app integrating LLM-powered chat for real-time, context-aware meditation coaching
• Result: Delivers personalized meditation sessions with natural conversation flow and real-time AI responses