I turn business challenges into data-driven, scalable solutions that create measurable business impact.

Financial institutions often misclassify borrower risk, leading to higher default rates and inequitable lending decisions.

Developed and compared multiple machine learning algorithms on a large scale credit dataset of 26K+ records. Implemented advanced preprocessing, feature engineering and applied hyperparameter tuning and ensemble methods to maximize predictive performance.

Achieved 97.7% accuracy, F1-score of 0.97, and AUC-ROC of 0.998, significantly improving recall for at-risk borrowers. The models also surfaced income and occupation trends, providing insights for fairer and more robust credit scoring.

Ad banner creation is slow, manual, and expensive for small businesses

Designed modular AI pipeline with GPT-3.5 for slogans, OpenCV for layout, and automated color/contrast rendering for production-ready banners

Cut banner creation time by 90%, enabling scalable, low-cost digital marketing

Premature mortality varies across 3,000+ US counties, but the key social, behavioral, and health drivers are not well understood.

Conducted exploratory data analysis (EDA) on socioeconomic, health, and behavioral datasets. Used regression and decision trees to model mortality outcomes and rank predictors.

Provided interpretable insights into top mortality predictors, supporting targeted interventions and public health planning.

Scaling AI captioning systems to handle heavy traffic while ensuring reliability and low latency

Deployed a serverless app using AWS (EC2, Lambda, S3, RDS) with Gemini AI for real-time auto-captioning, supported by secure architecture (IAM, subnets, Bastion)

Reduced annotation time by 50%, validated through high-load stress testing

Traditional malware scanners often work offline or rely on signature databases, making them too slow to detect zero-day or fast-evolving attacks.

Built a real-time malware detection system that monitors processes and network activity continuously, extracting behavioral features on the fly. Applied supervised machine learning models to classify processes as benign or malicious in real time, enabling proactive response instead of reactive scanning.

Demonstrated faster detection and reduced false negatives compared to baseline approaches, showing how AI-driven real-time monitoring can strengthen system security and resilience.

Training self-driving algorithms in real-world conditions is expensive and unsafe

Built a CNN-based car simulation with deep learning and computer vision, training models to navigate virtual driving environments

Successfully simulated lane detection and navigation in complex tracks, reducing reliance on costly real-world testing