Cody
软件工程 · 关注 AI / LLM / Agent
Software Engineering · AI / LLM / Agents
关于
About
软件工程背景,写过 Java 后端,也做前端;目前在内蒙古大学读软件工程硕士。
现在把重心放在 AI 与软件工程的交叉处:智能体系统、检索增强生成, 以及大模型正在如何改变软件本身的构建方式。
I come from a software engineering background — I've built Java backends and front-end interfaces. I'm currently a master's student in Software Engineering at Inner Mongolia University.
My focus sits where AI meets software engineering: agent systems, retrieval-augmented generation, and how large models are changing the way software gets built.
教育背景
Education
-
硕士研究生 · 软件工程(全日制)内蒙古大学
-
本科 · 计算机科学与技术(工学学士)内蒙古师范大学
-
Master's · Software EngineeringInner Mongolia University
-
Bachelor's · Computer ScienceInner Mongolia Normal University
动态
News
- 开源 codex-context-continuity,为长时间运行的编码任务保存上下文。
- 开源 flight-deals-mcp,一个本地 MCP Server。
- 开始记录技术笔记,最初写的是 SQL 与 Git。
- Released codex-context-continuity, keeping context alive across long coding tasks.
- Released flight-deals-mcp, a local MCP server.
- Started keeping technical notes — SQL and Git first.
方向
Research
- Agent
- 让模型能持续干活:上下文保持、工具调用、任务交接。
- RAG
- 检索质量与答案可溯源性,而不是只把向量库接上去。
- AI × SE
- 代码生成与评测:模型在实际工程流程里到底靠不靠谱。
- Agents
- Making models work over time: context, tool use, task handoff.
- RAG
- Retrieval quality and answer traceability, not just a vector store.
- AI × SE
- Code generation and evaluation — what actually holds up in practice.
项目
Selected Work
flight-deals-mcp
2026本地运行的 MCP Server,在国内航班数据上做发现与二次核验,让助手能直接查航班与价格。
给长时间运行的编码任务保存“工作卡片”,通过 GitHub 显式交接上下文,避免会话中断后从头再来。
flight-deals-mcp
2026A local MCP server that discovers and re-verifies domestic flight options, so an agent can search fares directly.
Durable work cards for long-running coding tasks, with an explicit GitHub handoff so context survives an interrupted session.
正在做的事
进行中这里放第三项:论文、实习项目或正在写的工具都行,两句话说明它解决什么问题。
In progress
OngoingAnything you're working on: a paper, an internship project, or a tool in progress.
写作
Writing
Notes from learning Java in 2020, kept for the archive.