Projects
Things I've built
Tools and systems focused on ML robustness, interpretability, and real-time applications.
Failure-First Model Stress Testing Tool
Built a failure-first evaluation framework to stress-test deep learning models under noise, occlusion, and distribution shift. Implemented degradation metrics that measure robustness and graceful failure rather than peak performance.
Interpretable ML Debugging Dashboard
Developed an interactive ML debugging dashboard for visualizing predictions, confidence drift, and attribution maps in real time. Integrated explainability methods to help identify model overconfidence and spurious correlations.
Philosophical Assumption Mapper
Built a tool to automatically identify and categorize implicit philosophical assumptions (ontological, epistemic, normative) in academic and technical texts using NLP-based analysis.
Real-Time Video Conferencing Application
Designed and implemented a real-time communication platform using Node.js and WebRTC for backend streaming. Built a React-based frontend interface and deployed the app on Vercel with seamless auto-scaling.