At IncodeVision, I developed Signal & Stay, an end-to-end churn intelligence workspace using the IBM Telco Customer Churn dataset. I benchmarked classification models with five-fold stratified cross-validation and used recall-oriented selection with F2-based threshold tuning.
I integrated a FastAPI prediction service with a React retention analytics dashboard, including risk scoring and explainable predictions. I also developed a CNN-based deepfake detection model that achieved 88% validation accuracy.
At TechBRJ, I developed a real-time feedback form in React.js during a two-week internship project. I’ve also built projects including ECHOEVER, a local-first AI memory archive, and ReCode, a DSA revision platform.

