At CNCF Meshery, I reviewed community pull requests and validated the Connections framework across Kubernetes, Prometheus, and Grafana. As a Linux Foundation Mentee, I implemented 21 pull requests across Meshery’s React/TypeScript frontend and Go backend, and worked with maintainers to unify three connection entry points into one reusable wizard.
For STAR, I led the methodology, software, and compute for a token-selection scheme fine-tuning Qwen3-8B with LoRA. As lead author, I reported lower held-out recovery-span loss while supervising 48.5% fewer tokens at equal wall-clock.
I also fine-tuned a Qwen2.5-Coder model with QLoRA for vulnerability detection and deployed an offline Ollama model that flags and patches OWASP flaws. My projects include NaviSense, an offline visual assistance app that won Hack4Innovation 2026, and an AI mock interview platform built with Next.js, FastAPI, SQLite, Gemini, and VAPI.
