Python, Flask, PostgreSQL, SQLAlchemy, Base62, Docker
- Built a production-grade URL Shortener and Analytics platform featuring custom aliases, expiration controls, and real-time click tracking across demographics and referrers.
- Implemented a Base62 algorithm for compact URL encoding and engineered a robust relational database schema with Flask and PostgreSQL.
XGBoost, Scikit-learn, Pandas, Streamlit
- Built an XGBoost-based fraud risk dashboard handling severe class imbalance (~6% fraud prevalence) using cost-sensitive learning, expanding 29 base features to 86 model features.
- Implemented business-centric evaluation (Audit Efficiency/Lift) achieving a 2.7x efficiency multiplier over random audits, with a real-time risk threshold slider deployed via Streamlit.
K-Means, DBSCAN, Hierarchical Clustering, Streamlit, Plotly
- Built an interactive mall customer segmentation dashboard using K-Means, DBSCAN, and Hierarchical Clustering with dynamic hyperparameter tuning and real-time cluster updates via Streamlit and Plotly.
- Implemented a supervised KNN proxy wrapper to enable out-of-sample predictions for DBSCAN and Agglomerative Clustering, with bulk CSV batch segmentation and single customer predictor tool.
CNN, TensorFlow/Keras, Flask, Python
- Built a CNN-based image classifier using TensorFlow/Keras with MaxPooling and Softmax activation to segregate waste into organic and recyclable categories.
- Deployed as a real-time web app using Flask with HTML/CSS frontend and hosted on Hugging Face Spaces.
Random Forest, Scikit-learn, Streamlit, Python
- Developed a heart attack risk predictor using Random Forest Classifier trained on clinical health metrics; performed EDA, feature selection, and 80/20 train-test split with Pickle serialisation.
- Built and deployed an interactive Streamlit web app enabling users to input health parameters and receive instant risk predictions.