AI systems and backend engineer based in Seattle. I am interested in agent runtimes, LLM inference infrastructure, developer tools, and production-oriented backend systems.
I learn by building small, inspectable systems: explicit interfaces, bounded execution, reproducible benchmarks, and tests around failure modes.
| Project | Focus |
|---|---|
| agent-runtime-lab | Minimal, testable ReAct-style runtime with typed model decisions, safe tools, timeouts, and JSONL traces. |
| inferops-agent | Benchmark-driven vLLM inference tuning agent for repeatable performance experiments. |
| mini-swe-agent | Source-reading and learning edition of a compact software-engineering agent. |
| rec-mlops | Recommendation-serving MLOps pipeline and operational workflow. |
| cpp-url-shortener | Production-style C++ short URL service and backend systems practice. |
| us-visa-scheduler | Open-source appointment monitoring with alerts and guarded automation. |
- gemini-cli: coding-agent architecture and CLI source reading.
- hmdp: Redis patterns and Spring Boot backend engineering.
- micrograd: scalar autograd, computation graphs, and neural-network fundamentals.
我是一名在 Seattle 的 AI 系统与后端工程师，关注 Agent Runtime、LLM 推理基础设施、开发者工具和生产级后端系统。
我的学习方式是构建小而清晰、可验证的系统：明确接口和边界，为执行设置超时，通过可复现 benchmark 衡量效果，并针对失败路径编写测试。
- Agent Runtime： 模型决策、工具调用、事件轨迹、超时与错误恢复。
- AI Infrastructure： vLLM 推理性能、benchmark、MLOps 与模型服务。
- Backend Systems： C++、Python、Redis、Spring Boot 和自动化工具。
上方的 Selected projects 是目前最适合快速了解我工程方向的项目。