I build open-source systems around AI agents, automated research, distributed computing, and data infrastructure. My work focuses on turning promising ideas into explicit architectures, reproducible experiments, observable runtimes, and deployable engineering systems.
我主要构建 AI Agent、自动化研究、分布式计算与数据基础设施 方向的开源项目，关注如何把有价值的研究想法转化为边界清晰的架构、可复现的实验、可观测的运行时和能够实际部署的工程系统。
- Agent infrastructure — control planes, runtime orchestration, safety boundaries, traces, and trainable telemetry
Agent 基础设施 —— 控制平面、运行时编排、安全边界、执行轨迹与可训练数据 - Automated research — repeatable loops for model, prompt, data, and strategy iteration
自动化研究 —— 面向模型、Prompt、数据和策略迭代的可复现研究闭环 - Distributed systems — algorithm runtimes, scheduling, checkpointing, messaging, and recovery
分布式系统 —— 算法运行时、调度、检查点、消息传递与故障恢复 - Data infrastructure — contract-first collection, validation, streaming, and downstream data products
数据基础设施 —— 契约优先的数据采集、校验、流处理与下游数据产品
| Project | Description |
|---|---|
| CareCopilot | A safety-aware medical-assistant agent backend with orchestration, auditable traces, and trainable telemetry. 面向中文医疗助手场景的 Agent 后端，强调任务编排、安全边界、可审计轨迹与训练数据沉淀。 |
| RAM-AutoResearch | An AutoResearch platform for recommendation, advertising, and marketing value optimization. 面向推荐、广告与营销价值优化的自动化研究平台，将实验、评估、分析与报告组织为可复现闭环。 |
| DistAlgo | A distributed algorithm execution framework for graph mining and machine learning. 面向图挖掘与机器学习的分布式算法执行框架，覆盖分区、迭代、消息传递、检查点与运行时适配。 |
| PromptHelix | Feedback-driven prompt optimization through closed-loop textual evolution. 通过评估、文本反馈、候选探索和策略演化实现闭环 Prompt 优化。 |
- AgentPlane — Control plane for AI infrastructure and agent runtime workloads.
面向 AI 基础设施与 Agent 运行时任务的控制平面。 - OpenEventFlow — Contract-first, end-to-end event data pipeline for large applications.
面向大型应用的契约优先、端到端事件数据管道。 - Meta-AutoData — A strict local implementation of the AutoData / Agentic Self-Instruct loop.
AutoData / Agentic Self-Instruct 数据挖掘闭环的本地严格实现。 - grpc-tfserving — A standalone C++ TensorFlow inference service with asynchronous gRPC serving.
基于异步 gRPC 的独立 C++ TensorFlow 推理服务。
Research ideas are only the beginning. I care about the systems that make them reproducible.
研究想法只是起点，我更关注让它们可复现、可验证、可运行的系统。
研究想法只是起点，我更关注让它们可复现、可验证、可运行的系统。