AI | Robotics | Dev Tooling | Ops
🐦 grackle · the control plane for AI coding agents
Configure once, supervise by exception — run Claude Code, Copilot, Codex, Goose & GenAIScript across Docker, SSH, Codespaces, and local from one CLI, web UI, and MCP server. Agent IPC, a Neo4j knowledge graph, and session resilience built in.
🎲 Reinforcement Learning for Yahtzee · A2C · PPO · REINFORCE
Teaching agents to play Yahtzee from scratch, benchmarked against an expectimax baseline — my UT Austin MSAI Case Studies capstone. Full research paper, training scripts, and W&B experiments included.
The always-on stack that runs my house: a self-hosted AI services bundle (ai-services — Ollama, Open WebUI, STT/TTS), a Rust orchestrator for isolated agent instances (zeroclaw-fleet), and battle-tested vLLM configs for running high-context agentic LLMs on a single RTX 3090 (local-llm-research). The house talks back through Claude as a Home Assistant conversation agent and a streaming TTS MCP server (pocket-tts-mcp).
| 🤖 AI & Agents | 🧠 Local AI & Homelab | 🛠️ Dev Tooling | 💼 Work @ Microsoft | ⚙️ Hardware & Robotics | 🎮 Apps & Fun |
|---|---|---|---|---|---|
|
|
|
|
|
|
|





