🚀 Software Engineer | Full-Stack, with a focus on AI/ML-backed products

_{This is a new account - my previous one was accidentally deleted. Older commit history and contributions are still attributed to me via my original commit email.}

I build complete, production-shaped systems end to end - backend APIs, databases, scheduling infrastructure, and frontends - with AI/ML woven in where it actually earns its place: agentic pipelines, structured LLM generation, and closed feedback loops, not bolted-on prompts.

• Build full-stack applications across the stack: FastAPI/Node.js/Spring Boot backends, React/Streamlit frontends, PostgreSQL/SQLite/Firebase persistence
• Build multi-agent LLM pipelines that discover, generate, review, and self-correct - hand-orchestrated directly on LLM APIs (Claude, Gemini, Groq) rather than black-box frameworks
• Design closed-loop systems that verify real-world outcomes and feed them back into future decisions (reflection/learning loops, not just static prompts)
• Build reliability layers around LLM output: structured/validated generation instead of free-text parsing, multi-stage automated review, reputation/trust scoring

Microsoft PyRIT (Python Risk Identification Tool for generative AI):

• #1831 - merged - added the SALT-NLP Moral Integrity Corpus (MIC) dataset loader _{Originally authored and merged from my previous GitHub account, which was accidentally deleted - the PR page shows "ghost" as a result, but the underlying commits are still attributed to me (verified via my original commit email).}
• #1910 - merged - fixed a temp-file leak and race condition in save_formatted_audio _{Originally authored and merged from my previous GitHub account, which was accidentally deleted - the PR page shows "ghost" as a result, but the underlying commits are still attributed to me (verified via my original commit email).}
• #2301 - merged - implements the Bijection Attack (arXiv:2410.01294) as a new converter/attack module