Building AI agents & LLM applications —
on top of years of high-performance systems engineering
These days I build AI agents and LLM-powered applications — tool orchestration, context engineering, and making agents reliable enough for real production use.
Before that I spent years building performance-critical backend systems: cloud-native infrastructure, high-throughput engines, in-memory platforms. That systems habit stuck — I still sweat the throughput, stability and clean architecture underneath every agent I ship.
Outside of work I test my limits in national programming contests. Every solution below is open-sourced and reproducible.
Contributor to openclaw — most recently keep proxied streams working when SSE data lines omit the space (merged).
| Result | Contest | Solution |
|---|---|---|
| 🥇 Champion | 11th CCF Big Data & Computational Intelligence Contest 2023 — graph pattern matching | geaflow-match |
| 🥈 2nd / 6506 | CNCF Cloud Native Challenge 2023 — Serverless cold-start elasticity | goscaler |
| 🥉 6th | OpenAtom Open Source Contest 2024 — sparse asymmetric banded linear systems (CUDA) | pbicgstab-ilu0 |
| 9th / 4404 | 1st Cloud Native Programming Challenge 2020 — distributed tracing | tianchi_2020_trace |
| 9th / 200 | KDD Cup 2024 — paper source tracing (NLP) | pst-kdd2024 |
AI — agent frameworks · LLM applications · RAG · prompt & context engineering Languages — Python · Java (concurrency, JVM) · Go · C++ / CUDA Systems — Kubernetes · RocketMQ / Kafka / MQTT · Spark / Hive
🇨🇳 中文版
构建 AI Agent 与 LLM 应用 ——多年高性能系统工程为底座
现阶段主线是 AI Agent 与 LLM 应用开发：工具编排、上下文工程， 以及让 Agent 在真实生产环境中足够可靠。
此前多年深耕性能敏感型后端系统：云原生基础设施、高吞吐引擎、内存级平台。 系统的习惯留了下来 —— 我依然会为每一个 Agent 背后的吞吐、稳定性与架构洁癖较真。
业余时间通过国家级编程竞赛检验技术上限，所有获奖方案均已开源、可复现。
openclaw 贡献者 —— 近期合入 修复代理流式 SSE 解析丢弃事件的问题。
| 成绩 | 赛事 | 方案 |
|---|---|---|
| 🥇 冠军 | 2023 第11届 CCF 大数据与计算智能大赛 · 图模式匹配 | geaflow-match |
| 🥈 亚军 (3/6506) | 2023 云原生编程挑战赛 · Serverless 冷启动弹性优化 | goscaler |
| 🥉 三等奖 (6/47) | 2024 开放原子开源大赛 · 非对称不定带宽线性方程组求解 (CUDA) | pbicgstab-ilu0 |
| Rank 9 (9/4404) | 2020 首届云原生编程挑战赛 · 分布式链路追踪 | tianchi_2020_trace |
| Rank 9 (9/200) | KDD Cup 2024 · 论文溯源 (NLP) | pst-kdd2024 |
AI — Agent 框架 · LLM 应用 · RAG · 提示词与上下文工程 语言 — Python · Java（并发 / JVM）· Go · C++ / CUDA 系统 — Kubernetes · RocketMQ / Kafka / MQTT · Spark / Hive
邮箱 pengzh1@qq.com —— 欢迎聊 Agent、LLM 应用，或一切够快、够有趣的东西。