鹿尧 Yao Lu

Experience

1. 高级软件工程师 Senior Software Engineer

   微软 (STCA · Bing / Skype / Copilot) Microsoft (STCA · Bing / Skype / Copilot) 2022-04 → 至今 → Present 苏州 Suzhou, China

   在 STCA WebXT/MAI 先后承担 Bing App / Start App、Skype、Copilot Mac 和 Surfaces Connectivity 的数据与质量工作；从业务数据 owner、指标体系和数据仓库，逐步扩展到跨端 telemetry、实验 scorecard、质量 RCA 和 AI-native 数据工作流建设。

   Owned data and quality work across Bing App / Start App, Skype, Copilot Mac, and Surfaces Connectivity in STCA WebXT / MAI. Started as the data owner for business metrics and warehouse foundations, then expanded into cross-surface telemetry, experiment scorecards, quality root-cause analysis, and AI-native data workflows.

   • 2025.11 - 2026.04 Cross-Surface Connectivity 质量分析

     转入 Surfaces Connectivity 后，负责 Mac / CMC 等跨端 no-response、actionable-rate 和 chat funnel 质量分析；为 Mac 对齐 Mobile 的 KQL、Grafana dashboard、Geneva monitor 和 baseline，并定位 CMC bot filter、background-tab 1DS terminal-event 丢失等 telemetry artifact，把“真实失败”和“数据采集失败”拆开，帮助团队聚焦正确 root cause。

   • 2025.02 - 2025.11 Copilot Mac 数据与质量体系

     在 Copilot Mac 数据基础薄弱的阶段，补齐核心指标、质量监控、实验 scorecard 和 feature dashboard；建立 ICM framework、daily / weekly OCV review 与 LT/PM 周会节奏，并主导 telemetry guidance、native rewrite backfill 和 voice funnel 分析，使留存、WAU、voice TDR 等核心指标能稳定跟踪、解释和优化。

   • 2022.04 - 2025.02 Bing App MiniApp / Growth / Acquisition 数据

     覆盖 Bing App MiniApp（Rewards、Wallpaper、Weather）以及后续 Growth / Acquisition 业务，负责核心指标、dashboard、funnel 和 ROI / LTV 分析；为 PM、运营和增长团队提供 feature 前后评估、Paid Ads 投放效率、Upsell funnel 与用户画像分析，支撑预算和增长策略决策。

   • 2022.04 - 2025.02 Skype Call Quality 与数据基础

     主导 Skype call quality funnel 建设，从 telemetry 梳理、funnel dashboard 到优化点识别和 impact 计算，支撑 1v1 call reachability / stability 等质量目标；同时对分散的数据表做关键模块数仓建模和 ODS / DWS 表开发，服务 PM dashboard、dev debug、AB scorecard 和质量 review。

   • Nov 2025 - Apr 2026 · Cross-Surface Connectivity quality analytics: moved into Surfaces Connectivity and owned no-response, actionable-rate, and chat-funnel analysis for Mac / CMC. Built Mobile-aligned KQL, Grafana dashboards, Geneva monitors, and baselines for Mac, and separated real product failures from telemetry artifacts such as CMC bot-filter issues and background-tab 1DS terminal-event loss.

   • Feb 2025 - Nov 2025 · Copilot Mac data and quality system: rebuilt core metrics, quality monitoring, experiment scorecards, and feature dashboards while the Mac data foundation was still thin. Set up the ICM framework, daily / weekly OCV reviews, and LT / PM review cadence; drove telemetry guidance, native-rewrite backfill, and voice-funnel analysis so retention, WAU, and Voice TDR could be tracked, explained, and improved reliably.

   • Apr 2022 - Feb 2025 · Bing App MiniApp / Growth / Acquisition data: covered Bing App MiniApp modules such as Rewards, Wallpaper, and Weather, then later Growth / Acquisition work. Owned core metrics, dashboards, funnels, and ROI / LTV analysis for PM, operations, and growth teams, supporting feature evaluation, Paid Ads efficiency, Upsell funnels, and user-profile analysis.

   • Apr 2022 - Feb 2025 · Skype call quality and data foundation: led the Skype call-quality funnel from telemetry cleanup and dashboard design to optimization-point discovery and impact sizing. Also modeled key warehouse modules and built ODS / DWS tables for PM dashboards, dev debugging, AB scorecards, and quality reviews.

2. 高级软件工程师 Senior Software Engineer

   小米 (大数据 · 用户画像 · MiPush) Xiaomi (Big Data · User Profile · MiPush) 2019-04 → 2022-03 北京 / 南京 Beijing & Nanjing, China

   在 MiPush 和用户画像方向负责 PB 级数据处理、实时/离线统计、竞品分析、DMP 标签和换机口径；作为 6 人数据小组的技术 Owner，除核心工程实现外，还承担任务拆解与排期、代码 review、招聘面试、新人培养和跨团队沟通。

   Worked on PB-scale processing, real-time and offline reporting, Android competitor analytics, DMP labels, and device-swap definitions across MiPush and user-profile systems. Served as technical owner in a six-person data team, covering core implementation, task breakdown, scheduling, code review, interviews, onboarding, and cross-team communication.

   • 2021.04 - 2022.03 用户画像 · 小米换机数据

     初期主导集团级换机口径定义、数据建模、代码实现和质量验证，解决多卡、多手、多设备流转下的序列识别问题，产出准确率 90%+、月活覆盖 80% 的换机标签；数据被全集团多个手机业务部门复用，用于品牌忠诚度、拉新画像和手机业务战略分析。

   • 2019.04 - 2021.04 MiPush · 安卓竞品数据

     接手并消化二十多个计算任务、数万行代码的竞品统计系统，基于 PB 级 MiPush 数据构建手机厂商新增与活跃指标；通过清洗、映射和 OAID 兼容，将与 BCI 的统计误差从 50%+ 拉回到 10%，并补齐 Kylin / Redash 可视化链路。

   • 2019.04 - 2021.04 MiPush · 手机销售赋能与 DMP 标签

     作为一期主程设计文案挖掘数据仓库和项目框架，开发三十余张表中的核心部分；构建购机意向打分和意向品牌标签，并为算法模型提供特征数据，支撑小米商城留存、去库存和多业务线精准投放。

   • 2019.04 - 2021.04 MiPush · 推送统计平台

     主导“米 Push 推送统计平台”重构，面向集团内部业务方和外部第三方开发者提供全选人数、推送量、接收量、点击量、CTR、分时接收/点击等核心报表；改造旧系统脉冲式写 HBase 和不合理数据结构，提升稳定性并减少存储空间，每年节省近百万元人民币成本。

   • Apr 2021 - Mar 2022 · User profile: Xiaomi device-swap data. Led the initial definition, modeling, implementation, and validation of a group-level device-swap taxonomy, handling multi-SIM, multi-owner, and multi-device-transfer cases. Delivered swap labels with 90%+ accuracy and 80% monthly-active coverage, later reused by multiple phone-business teams for loyalty, acquisition profiling, and phone-business strategy analysis.

   • Apr 2019 - Apr 2021 · MiPush Android competitor analytics: took over a legacy competitor-statistics system with 20+ compute jobs and tens of thousands of lines of code, then built vendor-level new and active device metrics on PB-scale MiPush data. Reduced error versus BCI from 50%+ to 10% through cleaning, mapping, and OAID compatibility work, and completed the Kylin / Redash visualization path.

   • Apr 2019 - Apr 2021 · MiPush sales enablement and DMP labels: as the main developer for phase one, designed the text-mining warehouse and project framework, implemented core parts of 30+ tables, built purchase-intent scores and intended-brand labels, and supplied feature data for algorithm models used in Mi Mall retention, inventory clearance, and precision marketing.

   • Apr 2019 - Apr 2021 · MiPush statistics platform: led the rewrite of the MiPush statistics platform for internal Xiaomi business teams and external developers, covering selected audience, sent, received, clicked, CTR, and hourly receive / click reports. Reworked old pulse-style HBase writes and poor data modeling, improving stability and reducing storage by nearly RMB 1M per year.

3. 大数据开发工程师 Big Data Engineer

   美图 (风控与反作弊) Meitu (Risk Control) 2018-04 → 2019-03 深圳 Shenzhen, China

   在社区风控与指标监控方向负责实时流处理、规则引擎、异常检测和告警平台工程化，把风控 PM / 运营规则配置、算法模型和生产告警串成可运营系统。

   Built real-time stream-processing, rule-engine, anomaly-detection, and alerting systems for community risk control and metric monitoring, turning PM rules, algorithm models, and production alerts into operational systems.

   • 2018.04 - 2019.03 风控实时规则引擎

     面向美图社区全链路风控，覆盖私信、关注、社区内容发布和文本创作等场景；基于 Spark Streaming + Kafka 将实时日志抽象成可查询内存表，用 QLExpress 支持风控 PM 自助配置规则，并接入文本相似、文本分类模型和类 SQL 因子，把原先偏离线、上线成本高的规则流程改造成低延迟实时引擎，显著缩短规则上线周期并降低人工成本。

   • 2018.04 - 2019.03 数据监控预警系统

     作为主要开发者参与公司级指标监控平台建设，服务美图内部数据团队的指标异常检测；接入数亿级指标并调用时间序列异常检测模型，异常命中准确率 98%+；负责 Spark 数据接入/重跑/查询、Redis 信号管理、Livy 服务化和邮件等告警链路的关键工程模块。

   • Apr 2018 - Mar 2019 · Real-time risk-control rule engine: built a low-latency review engine for Meitu community scenarios including private messages, follows, content publishing, and text creation. Used Spark Streaming + Kafka to expose real-time logs as queryable in-memory tables, QLExpress for self-service rule configuration by risk PMs, and text similarity / classification models plus SQL-like factors to shorten rule rollout cycles and reduce manual cost.

   • Apr 2018 - Mar 2019 · Data monitoring and alerting system: main developer on a company-wide metric monitoring platform for internal data teams. Ingested hundreds of millions of metrics and ran time-series anomaly models with 98%+ precision on detected anomalies; owned key Spark ingestion / rerun / query modules, Redis signal management, Livy service integration, and email alerting.

4. 大数据开发工程师 Big Data Engineer

   创维酷开 (数据平台) Skyworth Coocaa (Data Platform) 2016-07 → 2018-03 深圳 Shenzhen, China

   作为早期数据工程成员参与数据部从无到有建设，是多个模块的核心开发者 / Owner，覆盖 DMP、BI ETL、实时流处理、标签体系和数据挖掘能力。

   Early data-platform engineer and core developer / owner across DMP, BI ETL, streaming, tag systems, and data-mining capabilities as the data team was being built from scratch.

   • 2016.07 - 2018.03 DMP 平台

     作为从 0 到 1 核心开发之一，建设面向运营和营销团队的用户标签、人群圈选和画像分析能力；平台覆盖约 2000 万用户、近百个标签，使用 HyperLogLog 做人群规模快速估算，结合 Spark UDF / UDAF 计算画像指标，并通过任务优化将新增标签作业缩短约 1 小时。

   • 2016.07 - 2018.03 BI 平台与用户行为实时计算

     开发并维护用户播放行为表和多张播放相关 ETL 表，支撑数据分析师在 BI 平台产出面向管理层的报表；沉淀数十个常用 Hive UDF 提升分析效率，使用 Redis sorted set + Spark Streaming 计算智能电视端用户行为实时数据，并参与 K-means 用户聚类、协同过滤推荐等早期数据挖掘任务。

   • Jul 2016 - Mar 2018 · DMP platform: core 0-to-1 developer for user tags, audience selection, and profile analytics used by operations and marketing teams. The platform covered about 20M users and close to 100 tags, used HyperLogLog for fast audience sizing, Spark UDF / UDAF for profile metrics, and task optimization that shortened new tag jobs by about one hour.

   • Jul 2016 - Mar 2018 · BI platform and smart-TV behavior streaming: built and maintained playback behavior tables and related ETL for BI reports used by analysts and management. Developed dozens of Hive UDFs, used Redis sorted sets + Spark Streaming for real-time smart-TV user behavior computation, and worked on early data-mining tasks such as K-means clustering and collaborative filtering.

Capability Highlights

1. 大数据 Big Data

   数仓 / 实时计算 / 数据分析 / 指标体系 Data Warehouse / Real-Time Computing / Analytics / Metrics

   • 大数据工程底盘：十年经验覆盖数据平台、数据仓库、实时计算、用户画像、数据埋点 和数据分析；使用包括 Spark、Flink、Kafka、Hive、HBase、Redash、Grafana、Kusto、Geneva 等通用技术栈，以及微软内部的 Cosmos、Titan、Power BI 等数据工具。

     Data engineering foundation: 10 years across data platforms, warehouses, streaming systems, user profiling, event instrumentation, and analytics; hands-on with Spark, Flink, Kafka, Hive, HBase, Redash, Grafana, Kusto, Geneva, and Microsoft data tools such as Cosmos, Titan, and Power BI.

   • PB 级数据处理的 MiPush 数据系统：负责实时/离线推送统计、安卓竞品分析和面向内外部开发者的统计报表；接手竞品系统后，通过清洗、映射和 OAID 兼容，将与 BCI 的统计误差从 50%+ 拉回到 10%。

     PB-scale MiPush data systems: owned real-time / offline push statistics, Android competitor analytics, and reporting for both internal Xiaomi teams and external developers. After taking over the competitor system, reduced statistical error versus BCI from 50%+ to 10% through cleaning, mapping, and OAID compatibility work.

   • 数仓重构与成本优化：主导 MiPush 推送统计平台重构，把旧系统脉冲式写 HBase 和不合理数据结构改成更稳定的模型，减少 HBase 存储，每年节省近百万元人民币。

     Warehouse refactoring and cost reduction: led the MiPush statistics platform rewrite, replacing old pulse-style HBase writes and weak data modeling with a more stable structure. The new design reduced HBase storage and saved nearly RMB 1M per year.

   • 能影响业务决策的数据分析：为 Skype、Bing App、Growth / Acquisition、Copilot Mac 和 Connectivity 建 funnel、dashboard、AB scorecard、LTV / ROI 分析与数据驱动优化；典型结果包括 Skype 1v1 call 成功率 88% → 98%、Copilot Mac 新用户留存 24.0% → 31.7%、Voice TDR 46.3% → 31.93%。

     Business-facing analytics: built funnels, dashboards, AB scorecards, LTV / ROI analysis, and data-driven optimization loops for Skype, Bing App, Growth / Acquisition, Copilot Mac, and Connectivity. Examples include Skype 1v1 call success 88% → 98%, Copilot Mac new user retention 24.0% → 31.7%, and Voice TDR 46.3% → 31.93%.

   • 监控预警与数据质量体系：从美图公司级异常检测平台（异常命中准确率 98%+） 到 Microsoft Geneva / Grafana monitor 和 CMC 1DS telemetry-loss，能把生产问题拆成可追踪、可告警、可复盘的数据系统。

     Monitoring and data-quality systems: built systems that turn production issues into trackable, alertable, and reviewable data, from Meitu’s company-wide anomaly detection platform with 98%+ precision to Microsoft Geneva / Grafana monitors and CMC 1DS telemetry-loss RCA.

2. 全栈开发 Full-Stack Development

   云服务 / 客户端 / 内部工具 / 开源贡献 Cloud Services / Client / Internal Tools / Open Source

   • 数据密集型产品里的端到端工程能力：把原始 telemetry、云服务、dashboard、脚本和轻量产品界面串成 PM、工程师和数据同学能直接使用的工作流。

     End-to-end engineering in data-heavy products: connects raw telemetry, cloud services, dashboards, scripts, and lightweight product surfaces into workflows that PMs, engineers, and data partners can use directly.

   • Microsoft Azure 云、telemetry 与 Mac 开发经验：熟悉 Azure 云服务使用，负责过 Copilot client telemetry 埋点与指标设计；在 Copilot Mac 中参与 Mac feature 开发。

     Microsoft Azure, telemetry, and Mac development: worked with Azure services, Copilot client telemetry instrumentation and metric design, and Copilot Mac feature development.

   • 内部工具与 dashboard 开发：近两年持续开发 Python 为主、少量 JS / TS 的数据可视化、Kusto 查询助手、MCP 工具和数据查询 agent，主要服务自己和团队同学的日常分析、排障等。

     Internal tools and dashboards: over the past two years, built Python-heavy data visualizations, Kusto query helpers, MCP tools, and data-query agents for personal and team analysis / debugging workflows.

   • 开源项目里的全栈贡献：向 NousResearch/hermes-agent 合入超过 15 个 commit，覆盖 ACP、agent runtime、config、tools、Discord / Telegram / Desktop / Gateway；向 openclaw/openclaw 合入超过 8 个 PR，覆盖 messaging adapter、agent dispatch 和 transport。

     Full-stack open-source contributions: merged 15+ commits to NousResearch/hermes-agent across ACP, agent runtime, config, tools, Discord / Telegram / Desktop / Gateway, and 8+ PRs to openclaw/openclaw across messaging adapters, agent dispatch, and transport.

   • 自建产品实践：Fawn（育儿 Agent，暂未开源，包括服务端、Android 与 iOS App）、voice-buddy、code-while-shit、dashboard-gen-skill、golden-flower 等项目覆盖 CLI、前端、后端、agent 集成和 dashboard 交付；熟练通过 Vibe Coding 快速进入新技术栈并做出可用版本。

     Self-built product practice: built and iterated Fawn, a private parenting agent with server, Android, and iOS components, along with voice-buddy, code-while-shit, dashboard-gen-skill, golden-flower, and other projects covering CLI, frontend, backend, agent integration, and dashboard delivery. Comfortable using AI-assisted coding to enter a new stack quickly and ship a usable version.

3. AI Agent AI Agent Engineering

   RAG / MCP / AI Native Workflow RAG / MCP / AI-Native Workflow

   • FY25 Microsoft AI School China 一等奖：AI Agents 赛道独立提交 Copilot Mac Data Query Agent，从 Microsoft 中国 AI 培训项目多个赛道的决赛项目中胜出。

     1st Prize · FY25 Microsoft AI School China: submitted Copilot Mac Data Query Agent solo in the AI Agents track and won from the finalist projects across multiple Microsoft China AI training tracks.

   • Copilot Mac Data Agent：在没有成熟数仓、只有原始 telemetry 的情况下，构建 KG-RAG agent 提升 Kusto 查询效率；技术栈包括 OpenAI Agent SDK、MCP-style tool calling、Neo4j、Kusto connector、Azure Container Registry，核心难点是元数据理解、指标定义、业务理解等。

     Copilot Mac Data Agent: built a KG-RAG agent to improve Kusto query efficiency when only raw telemetry was available and no mature warehouse layer existed. The stack included OpenAI Agent SDK, MCP-style tool calling, Neo4j, a Kusto connector, and Azure Container Registry; the hard parts were metadata understanding, metric definition, and business context.

   • Agentic data workflow：把 KQL / SQL、dashboard 迭代、OCV 报告、AB 读数、single-user telemetry 调查等高频数据工作逐步迁移到 AI 辅助流程，用数据经验约束 agent 输出质量。

     Agentic data workflow: gradually moved recurring data work such as KQL / SQL drafting, dashboard iteration, OCV reports, AB readouts, and single-user telemetry investigation into AI-assisted workflows, using data experience to constrain agent output quality.

   • Agent 源码与开源实践：独立维护 Claude-Code-Source-Study（1.5K+ stars / 530+ forks），并把源码阅读沉淀转化为 Hermes Agent、OpenClaw 等项目的真实合入贡献。

     Agent source-code study and open-source practice: independently maintains Claude-Code-Source-Study (1.5K+ stars / 530+ forks) and turned that source-reading practice into real merged contributions to Hermes Agent and OpenClaw.

   • AI 辅助复杂排障：通过 LLM 多轮辅助阅读 Edge Chromium 与 Picasso 等 40GB+ 规模的源码，支撑 CMC bot filter 和 terminal-event 丢失 RCA，把 agent 能力用于生产级开发和 debug，而不只是简单的 demo。

     AI-assisted debugging on large codebases: used multi-pass LLM-assisted reading across 40GB+ of Edge Chromium and Picasso source to support CMC bot-filter and terminal-event-loss RCA, applying agent workflows to production engineering and debugging rather than simple demos.

Open Source Projects

• Claude-Code-Source-Study Claude-Code-Source-Study

  1.5k+ stars / 536+ forks 1.5k+ stars / 536+ forks

  个人源码研究与学习项目，围绕Anthropic在2026年4月“开源”的Claude Code版本的代码，做系统性源码解读与学习。中文社区影响力 最高的 Claude Code 内核研究参考工具书籍之一。

  Personal source-code study project based on the Claude Code codebase released by Anthropic in Apr 2026. It is one of the most visible Chinese references for understanding Claude Code internals.

• Hermes Agent Hermes Agent

  15 个 main commit · ACP / agent runtime / config / tools / multi-channel 15 main commits · ACP / agent runtime / config / tools / multi-channel

  向 NousResearch/hermes-agent main 分支合入贡献，覆盖 ACP session、agent runtime、配置校验、工具调用，以及 Discord / Telegram / Desktop / Gateway 等多 channel 集成问题。

  Merged contributions to NousResearch/hermes-agent main across ACP sessions, agent runtime, config validation, tool calling, and Discord / Telegram / Desktop / Gateway integrations.

• OpenClaw OpenClaw

  8 个 merged PR · messaging adapter / agent dispatch / transport 8 merged PRs · messaging adapters / agent dispatch / transport

  向 openclaw/openclaw 合入多项修复，集中在 GoogleChat / Matrix adapter、active memory deadline、ACP dispatch path、OpenRouter transport、embedded session takeover 等真实工程问题。

  Merged fixes to openclaw/openclaw across GoogleChat / Matrix adapters, active-memory deadlines, ACP dispatch, OpenRouter transport, and embedded session-takeover paths.

• AI-Native Productivity Tools AI-Native Productivity Tools

  Python / TypeScript / HTML · CLI / dashboard / agent workflow Python / TypeScript / HTML · CLI / dashboard / agent workflow

  围绕AI coding workflow 持续做工具实验，包括 voice-buddy、dashboard-gen-skill、skila、watch-claw、golden-flower 等；重点是把想法快速推进到可运行、可复用的小工具。

  Tool experiments around AI coding workflows, including voice-buddy, dashboard-gen-skill, skila, watch-claw, and golden-flower; the goal is to move ideas quickly into tools that run and can be reused.

• GitHub 公开工程信号 GitHub public engineering record

  源码研究 / AI Agent 工具 / 数据可视化 / CLI / 小型 Web 应用 source study / AI-agent tools / data visualization / CLIs / lightweight web apps

  公开项目覆盖源码研究、AI Agent 工具、数据可视化、CLI、小型 Web 应用和游戏实验；主要公开语言包括 Python、TypeScript、JavaScript、HTML、Shell、C# 和 GDScript。

  Public work spans source-code study, AI-agent tools, data visualization, CLIs, lightweight web apps, and game experiments. Main public languages include Python, TypeScript, JavaScript, HTML, Shell, C#, and GDScript.

Skills

Awards

• 一等奖 · FY25 AI School China 1st Prize · FY25 AI School China

  Microsoft Microsoft · 2026-02

  AI Agents 赛道独立提交 Copilot Mac Data Agent (KG-RAG)，并在 Microsoft 中国 AI 培训项目决赛中胜出；决赛项目覆盖 AI Agents / RAG / Fine-tuning / Multimodal 等方向。

  Won the AI Agents track with a solo submission, Copilot Mac Data Agent (KG-RAG), selected from the Microsoft China finalist pool across AI Agents, RAG, Fine-tuning, and Multimodal tracks.

• Transformational Impact · FY25 最高等级绩效 Transformational Impact · Highest FY25 Performance Outcome

  Microsoft Microsoft · 2025

  因在 Copilot Mac 数据基础设施建设中的核心贡献获得 FY25 最高等级绩效评价；工作覆盖 telemetry、dashboard、scorecard 与质量 review 机制，成为团队产品质量和增长分析的重要基础。

  Received Microsoft’s highest performance outcome for FY25, with specific recognition for building Copilot Mac’s data foundation across telemetry, dashboards, scorecards, and quality review loops.

• 破格晋升16级 Consecutive A Ratings · Accelerated L15 to L16 Promotion

  小米 Xiaomi · 2019-2021

  因小米换机数据和 Android 竞品数据体系建设连续获得 A 级绩效；一年内从 15 级晋升至 16 级，快于常规晋升节奏。

  Earned consecutive A-level performance ratings for Xiaomi device- swap data and Android competitor analytics work, and was promoted from L15 to L16 in about one year, ahead of the normal promotion cadence.

Education

• 湖南大学 Hunan University (985)

  985 · 信息与计算科学 B.S. · Information and Computing Science · 2012-09 → 2016-06 · 长沙 Changsha, China

Languages

• 中文 Chinese — 母语 Native
• 英文 English — 工作流利（日常技术读写） Professional working proficiency (technical reading & writing daily)