Building infrastructure for AI agents — MCP tools, plugin ecosystems, agent orchestration, and distributed training.
- 🧠 Building an MCP ecosystem for text-only coding agents — vision, memory, and tooling that lets DeepSeek-class models see and act.
- 🤖 Exploring agent orchestration — multi-agent systems, plugin marketplaces, autonomous PR workflows.
- ⚡ Deepening distributed training — Megatron-LM, ZeRO, and inference optimization.
- 👁️ mcp-vision-bridge — Give text-only LLM agents vision: an MCP server that routes images through any multimodal model (mimo, Claude, Gemini). Plugin + one-command installer, works across Claude Code / Codex / opencode.
TypeScript · MCP - 📦 agent-plugins — A Claude Code plugin marketplace: install agent extensions in one command.
Marketplace - 🔁 Auto-PR Workflow — AI agents that autonomously submit high-quality PRs.
Python · Claude Code - 🐝 OmniHive — Multi-agent orchestration: task decomposition, discovery, collaboration.
Rust · Tauri - 📈 Sextant — AI-native quant engine: agents read market state via mmap, output structured intents.
Rust · NautilusTrader - 🧪 AgentMind — Multi-core independent evaluation with context isolation for AI agents.
Python
- 🏗️ Infrasys-AI/AIInfra — 11 PRs: Megatron-LM, ZeRO memory optimization, TP/PP, MoE, OpenCompass.
- 🔧 anthropics/claude-agent-sdk-python — Windows CLI path handling, test fixes.
- ⚙️ warpdotdev/warp — Hermes CLI agent detection.
35+ merged PRs total across 15+ projects.


