• Inference Serving Stack: Scheduling and routing across heterogeneous models — the production path for vLLM and llm-d workloads on Kubernetes.
• Kubernetes-Native Workload Plumbing: Batch queueing, multi-cluster scheduling, and GPU sharing for AI/ML workloads — contributing upstream to Kueue, Karmada, and HAMi.
• Agent Harnesses: The supervision layer for long-running LLM agents — parallel sessions, multi-agent teams, artifacts, approval gates, scheduled dispatch. Because the work shouldn't collapse into chat.
• Agentic Developer Workflows: Multi-worktree dispatch, Claude Code skills, and commit/PR/task automation — the inner loop I live in daily.

• OpenClaw (Core Contributor, Current Focus): Open-source agent infrastructure for long-running, multi-channel AI work.
• semantic-router (Committer): Defining the decision-making layer for multi-model LLM serving.
• llm-d: Cloud-native infrastructure for disaggregated LLM inference.
• HAMi & Kueue: Kubernetes-native batch scheduling and GPU virtualization.
• Istio: Traffic governance for the service mesh layer.
• Karmada & Kubernetes: Multi-cluster orchestration foundations.

I build tools to fix my own problems.

• ClawWork: Client for OpenClaw — connect ClawWork to your own OpenClaw and unlock multi-session productivity.
• lathe: Agent-friendly CLI generator for APIs: turn Swagger, OpenAPI, and google.api.http protos into single-binary Cobra CLIs with catalogs and generated Skills.
• awire: Cloudflare Worker control plane — route source events (GitHub) to delivery channels (Feishu) as interactive cards. Pure KV, HTMX admin, no build step.
• agent-brains: Portable domain agent brains for coding agents — AGENTS.md, SOUL.md, MEMORY.md, skills, workflows, and a bundled loader.
• Recall: Local-first search across 10+ AI coding tools (Claude Code, Codex, OpenCode, Cursor, Gemini, Cline, Pi, Kiro, Copilot CLI, Antigravity) — hybrid FTS + embeddings, usage dashboard, all on-device.
• gmc: Parallel git worktrees for parallel AI agents, plus AI-generated commits.
• merge-scout: A Claude Code skill for vibe coding — ranks GitHub issues by contributability × merge probability so your agent picks work that will actually land.
• codex-agents-local: Local AGENTS.local.md overlays for Codex via hooks, without replacing the official codex command.
• pared: Minimal native macOS video compressor — drop a file, pick a style, done.
• mailbell: Minimal macOS menu bar notifier for Gmail.
• mote: Menu bar macOS app for rewriting selected text with OpenAI-compatible models.
• branchlight: Checking PRs in browser tabs is a ritual. Moved it to the menubar.
• MacMusicPlayer: A minimalist, clean music player for macOS.
• SaveEye: A minimalist eye care reminder that doesn't annoy you.
• Chrome TabBoost: Browser tab overload is a bug. I patched it with an extension.
• ConfigForge: Manage ~/.ssh/config and kubeconfig without the headache.
• LogoWallpaper: Generating brand assets shouldn't take 30 minutes.
• prooflet: The proof layer for prototypes — in-page narration and annotation SDK, no backend in V1.
• mdctl: AI-powered Markdown workflow automation.
• hf-model-downloader: Painless Hugging Face model downloads.
• mirrormate: Docker pulls failing? I fixed it with magic.
• swagger-online: Unified Swagger UI. No more tab chaos.
• ai-icon-generator: I needed icons, so I built a generator.
• gofs: python -m http.server is slow. Rewrote it in Go.
• gh-x: Batch repo operations as a gh CLI extension. One-by-one doesn't scale.
• convostore: Stateful LLM APIs shouldn't be this hard. A Redis-backed conversation store for OpenAI Responses API / vLLM.
• prompts: Prompt engineering is the real frontend of LLMs. I track mine here.
• openclaw-gateway-tunnel (RFC, design phase): Zero-binary ngrok tunnel for OpenClaw gateways. Looking for co-designers before code lands.
• moltbot-channel-feishu (legacy, superseded by OpenClaw official): Feishu/Lark channel plugin for Moltbot/Clawdbot.