I build AI systems that move from model output into controlled action: robot autonomy harnesses, Unreal/3D-world control surfaces, and memory/control layers for LLM agents. My public repos also include edge-oriented LLM deployment work and neural-architecture experiments; the best work here is prototype-heavy and documented with explicit safety or validation boundaries rather than product claims.

• pip-vector-autonomy - FastAPI brain service for an Anki Vector robot with 25 Python modules and a deterministic safety filter between LLM action plans and SDK execution.
• UEAgentForge - Unreal Engine 5 editor-control bridge; .uplugin reports version 0.5.0, and the Python client exposes 122 methods for Remote Control/MCP workflows.
• HorizonEngine - TypeScript/WebGPU engine prototype with 17 workspace packages, five example apps, and 15 .test.ts files across engine and AI-control-plane subsystems.
• Thought - LLM thought-tagging and memory-system prototype; python -m pytest -q passed 62 tests locally on 2026-05-28.
• Aero + AeroNum - Compiler and numerical-runtime experiments; Aero's public benchmark artifacts are tracked at commit baf5d7a, and AeroNum reports a verified HIP vector-add median of 0.259509 ms for 16,777,216 float32 elements on an RX 7900 XTX.
• Liquid-neural-network - PyTorch neural-architecture playground with liquid, attention, and multi-scale variants; pytest collection currently fails until missing dependencies/modules are fixed.

Not listed: forks, classroom repos, empty stubs, and template projects.