I build AI agents, automation tools, and runnable product prototypes for real-world workflows. My work sits at the intersection of agent orchestration, governed data access, and human-in-the-loop automation.
- AI agents & automation — tool-using LLM systems, workflow orchestration, and operational prototypes
- Governed data systems — auditable agent access, semantic layers, access controls, and reversible actions
- Applied ML & embodied AI — reinforcement learning, behavioral cloning, and interactive agents
- Product engineering — Python and TypeScript systems that can be run, inspected, and validated
- DataSteward — a self-hosted reference implementation for governed, AI-agent-native data access
- Cyber-Superego — a privacy-conscious, edge-cloud productivity supervision prototype
- Minecraft AI Companion — a dual-agent Minecraft companion built with LLM orchestration
- Minecraft Embodied AI Agent — behavioral cloning and PPO experiments on Apple Silicon
- ScholAI — a research prototype for meaning-preserving text transformation and evaluation
I start with a concrete workflow, make the smallest system that can be run and checked, then improve it with clear interfaces, evaluation, auditability, and explicit safety boundaries.
This profile links only to public work. Project repositories contain their own setup, architecture, and status documentation.