Brave Buddy is a voice-based AI assistant designed to support children facing cancer. Using GPT-4o, Whisper AI, and NeMo Guardrails, it enables emotionally safe, age-appropriate conversations — from reminders to medical Q&A. Built by a cross-functional team, it won 2nd place at the Rutgers Health Hackathon, blending empathy with real-time AI to ease pre-surgical anxiety and offer a personalised support experience.

Built a real-time crypto forecasting model that predicts short-term Bitcoin price movement using sentiment and persuasion signals from Reddit and news headlines. Engineered temporal NLP features with FinBERT and GPT-4o, achieving a 12% ROC-AUC improvement over prior benchmarks. Designed for use in high-volatility trading environments to support risk-aware decisions.

A modular machine learning pipeline for multi-class cancer detection built with TensorFlow, MLflow, and Docker. Achieving 96% accuracy on histopathological images, this project goes beyond model training to deliver version-controlled workflows, experiment tracking, and a deployable inference UI; all designed for reusability in real-world medical imaging environments.

CrashAlert is a predictive system that uses road, weather, and traffic data to identify high-risk accident scenarios and suggest nearby hospitals for first aid. Built using PySpark and Spark MLlib on over 7 million records, it forecasts crashes with 94.2% accuracy and proposes emergency responses — designed for future integration into smart vehicle systems.

A no-code AutoML assistant powered by GPT-3.5 and Scikit-learn that lets users build and tune ML models just by chatting. From setting up training workflows to getting classification reports and confusion matrices, everything happens through simple prompts. Designed to make machine learning more accessible, visual, and—dare we say—enjoyable.