Forest Fire Prediction

Developed a two-stage pipeline using K-Means clustering and Random Forest to classify wildfire risk into five tiers from ~100k hourly weather records, achieving ~100% accuracy across metrics and validating with cross-validation, confusion matrices, and silhouette scores.

Face Recognition using Eigenface & Naive Bayes

Implemented a PCA-based Eigenface and Naive Bayes pipeline with z-score normalization, boosting accuracy from 27% to 88% via 10-fold cross-validation, hyperparameter tuning, and visual evaluations with heatmaps and classification reports

Fooducate (Capstone Project)

Built a React/Node nutrition app with Firebase real-time data and an OpenAI-powered chatbot for instant diet tracking and guidance.

Cyber Attacks on LLM (Research Survey)

Researched LLM security: uncovered model inversion risks, forecasted a 25% rise in prompt-injection attacks, and outlined gaps in current defenses with universal mitigation proposals.