8000
Skip to content
View tytsxai's full-sized avatar

Block or report tytsxai

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
tytsxai/README.md

qilai — AI application and delivery engineer. Based in China, GMT+8. Four years of engineering, AI in production since 2023. Available immediately, onsite or remote.

研发交付工程师 / AI 落地交付 · Node.js · TypeScript · Python · React
把 AI 接进真实业务流程,做成能上线、能排障、能交给别人跑下去的系统。

wwtvn1937@gmail.com · 全部仓库


WHAT

能力方向

常驻中国,时区 GMT+8。工程经验约四年(2022.03 起),2023 年起把 AI 工程化落地到生产环境,目前可随时到岗,远程或驻场都行。

核心方向是把模型接到扛得住生产的地方:Agent 工作流、Tool Calling、结构化输出,MCP / RAG 以接入落地为主,不做算法研究;失败可重试、过程可回查、异常留有人工接管路径。主要积累在异步系统的状态机与幂等设计——模型 API、三方平台、链上接口这类外部依赖,通常做成可重试、可回放、可排障的模块,涉及状态一致性与补偿审计。检索方向做过全文 + 向量混合召回、分数融合与 rerank、中文分词与相似去重,后续的 RAG 知识库检索层沿用了这条链路。

交付覆盖接口、库表、React 页面、Docker 部署、GPU 环境上的模型服务与线上排障,通常一人负责到底。节奏上先用一到两周产出可用版本,再补齐幂等、重试、人工接管与回滚,随交付附带验收清单、Runbook 和部署脚本。

公司内部项目包括 AI 自动化运营平台、内容生产线、AI 图像处理平台与中文混合检索系统;源码不对外公开,架构与技术选型可在沟通中展开。

English

AI application & delivery engineer, based in China (GMT+8). Roughly four years of engineering since 2022.03, AI in production since 2023. Available immediately, onsite or remote.

Core focus: putting models where they have to survive production — agent workflows, tool calling, structured output, MCP and RAG as integration work rather than research — with retries, replayable state, and a human handoff for anything that can get stuck. Primary strength is async state machines and idempotency: model APIs, third-party platforms and chain interfaces turned into modules that can be retried, replayed and debugged. On the retrieval side: full-text + vector hybrid recall, score fusion, rerank; a later RAG pipeline reused the same recall layer.

Delivery covers API and schema design, the React front end, Docker deployment, GPU model serving, and on-call debugging — usually end to end. Typical pace: a usable version in a week or two, then idempotency, retries, human takeover and rollback, shipped with an acceptance checklist, a runbook and deploy scripts. In-house work has included an AI automation platform, a content pipeline, an image-processing platform and a Chinese hybrid search system; source stays private, architecture is open to discuss.

Contact: wwtvn1937@gmail.com — the only channel on this profile.


TOOLS

Stack: TypeScript, Python, Swift, Rust, Bash; LLM APIs, tool calling, agent workflows, JSON Schema, MCP, RAG; Node.js, FastAPI, Express, PostgreSQL, Redis, BullMQ; React, Next.js, Vite, Tailwind, SwiftUI; Meilisearch, pgvector, hybrid recall, rerank, SimHash dedup; Docker, Linux, Nginx, GitHub Actions, GPU model serving; idempotent retries, human handoff, structured logs, cost per call

Kubernetes 是线上运维的参与深度,不是主栈,图里没单列出来。


EVIDENCE

能点开的证据

公开、可运行、在维护。星标数以仓库页为准,正文不手写数字。

仓库 一句话
IDM-Activation-Script-Chinese Windows 批处理工具集 · GPL-3.0。有真实用户量的分发与维护:issue 承接、GBK / CP936 编码坑、注册表自动备份、失败可回退
cleanplate 本地优先的 AI 视频去水印后端 · FastAPI + ComfyUI DiffuEraser,FFmpeg 降级。异步 job 架构与模型服务化的公开版本
PromptPanel macOS 提示词与片段启动器 · Swift / SwiftUI,本地优先、全局快捷键。文档、CI、单测、发布与恢复流程全部走完
bilibili-cleaner 账号批量清理 · FastAPI + Web UI + CLI + Docker。破坏性自动化怎么做得住:扫码登录、限速、复核门、pytest CI
macfriends-cli macOS 微信好友关系检测 · Rust + ObjC++ agent。系统层深度:原生 agent、ABI 锁版逆向、纯本地不联网
bazi-master 全栈命理平台 · React + Express + PostgreSQL + Prisma。LLM 进产品:解读流水线、多模块领域逻辑、全栈自持

Selected work — pinned repositories with their primary language and live star count

Language mix across all public repositories, by bytes of code

上面每张图都由 scripts/gen_assets.py 从 GitHub API 生成、CI 每周刷新。
自建而不挂第三方卡片服务:那些服务限流之后会渲染成裂图,主页裂了传达的信息是错的。


CONTACT

联系

wwtvn1937@gmail.com —— 本主页唯一的联系方式。

邮件里说明要解决的问题即可,比发 JD 更有效;通常一到两封邮件内能确认是否能接、大致方案与所需前提。

Email wwtvn1937@gmail.com — the only contact channel here. A problem description works better than a job description.

AI 只有活过生产环境才算数;自动化只有还在控制之内才算数。

Pinned Loading

  1. PromptPanel PromptPanel Public

    Local-first macOS prompt/snippet launcher (global hotkey) for AI apps. macOS Prompt 启动器。

    Swift 4

  2. bilibili-cleaner bilibili-cleaner Public

    Bulk-clean Bilibili follows/favorites/dynamics/history — QR login + Web UI. B 站账号批量清理。

    Python 5

  3. bazi-master bazi-master Public

    Global divination web app: BaZi, Tarot, I Ching, Zodiac + AI (React/Express/Postgres). 命理全栈平台(八字/塔罗/易经/星座)。

    JavaScript 20 5

  4. virtual-chem-lab virtual-chem-lab Public

    Gamified virtual chemistry lab for teaching and simulation. 游戏化虚拟化学实验室。

    Python 2

  5. IDM-Activation-Script-Chinese IDM-Activation-Script-Chinese Public

    Chinese GBK toolkit for IDM trial/activation/reset with registry backup (does not patch IDM binary). IDM 激活脚本中文版。

    Batchfile 354 22

  6. cleanplate cleanplate Public

    Local-first AI video watermark removal API. FastAPI + ComfyUI DiffuEraser, FFmpeg fallback. 本地优先的 AI 视频去水印后端。

    Python

0