AI Agent & Evaluation Engineer · Indie Game Dev Explorer
AI Agent 与评测工程 · 独立游戏与图形计算探索者

• AI Agent Harness & Tooling — Building and extending plugin-first agent harnesses (DeepSeek Harness), MCP tool bridges, multi-agent workflows, System-1 probabilistic routing (Jev), and macOS developer utilities (CodexBar).
  Agent 框架与工具链：深耕全插件化 Agent 架构（DeepSeek Harness）、MCP 协议桥接、多智能体编排、确定性概率评估管线以及 macOS 开发者效率工具。
• Multimodal & Hardware-in-the-Loop Evaluation — Designing LLM evaluation platforms, voice/ASR intent benchmarks, and vision-driven Android GUI testing agents with dual-track (structured IR + isolated VLM judge) assertions.
  多模态与软硬一体评测：构建 AI 评测平台、语音/ASR 意图评测工具链，以及基于「结构化 IR + 独立视觉裁决」双轨机制的端侧 UI 自动化测试 Agent。
• Game Dev & Graphics — Prototyping HD-2D co-op RPGs in Godot 4 (GDScript) and experimenting with Rust (wgpu / WebAssembly) and foundational deep learning (micrograd / Transformers from scratch).
  游戏开发与图形计算：使用 Godot 4 研发 2D 俯视角合作 RPG，并持续探索 Rust 图形渲染（wgpu / WASM）与深度学习底层实现。

• Languages: TypeScript · Python · Rust · GDScript · Swift
• AI & Agents: DeepSeek Harness (Cordis) · MCP · LLM-as-a-Judge · Vision UI Agents · PyTorch
• Engine & Runtime: Godot 4 · Node.js · WebGPU / WASM · macOS & Arch Linux

• ✉️ Email: xz.ganbaruibro@gmail.com