// All resume content, bilingual. Pulled directly from the user's content. // Lang = 'zh' | 'en' const RESUME = { meta: { name: { zh: '孟依凡', en: 'Yifan Meng' }, target: { zh: '字节跳动 平台产品经理（AI方向）· TikTok直播', en: 'Platform PM (AI) at ByteDance · TikTok Live', }, location: { zh: '现居杭州', en: 'Based in Hangzhou' }, phone: '13226605632', email: 'aurora9716@qq.com', edu: { zh: '华南理工大学 · 信息与交互设计 · 本科 (2015–2019)', en: 'South China University of Technology · BA, Information & Interaction Design (2015–2019)', }, }, hero: { eyebrow: { zh: '在线产品 · 已上线运行', en: 'Live product · in production' }, title: { zh: 'Axiom 元溯', en: 'Axiom' }, subtitle: { zh: '个人天赋觉察 Agent 应用', en: 'A personal-talent discovery Agent', }, pitch: { zh: '独立设计与开发的 AI 对话产品。从对话工作流架构、知识库结构化、B 端后台到 trace 评估的全链路产品决策。', en: 'Solo-built AI conversation product. End-to-end product decisions across dialog workflow, knowledge base, admin tooling and trace evaluation.', }, stats: [ { v: '300+', l: { zh: '真实用户', en: 'real users' } }, { v: '12', l: { zh: '人均对话深度', en: 'avg. turns / user' } }, { v: '2.5', l: { zh: '天独立交付 v1', en: 'days · solo v1 delivery' } }, { v: '6', l: { zh: '节点对话工作流', en: 'node workflow' } }, ], link: 'axiom.nyota.cn', }, // SUMMARY — 个人优势 strengths: { title: { zh: '个人优势', en: 'Strengths' }, items: [ { tag: { zh: '6 年实战', en: '6 yrs in industry' }, head: { zh: '互联网头部产品实战经验', en: 'Top-tier Chinese internet product experience', }, body: { zh: '近 3 年聚焦双边开放平台与 B 端工具产品——从 0 到 1 开放平台 C 端运营建设、独立交付主播挂载工具开放平台、设计 T+1 自动化分发系统等多场景 B 端工具产品工作；具备将专业运营 SOP 沉淀为可配置工具包、推动核心指标达成的完整方法论。', en: 'Last three years focused on two-sided open platforms and B-end tooling — building 0→1 C-end ops for an open platform, solo-delivering an anchor-tool open platform, and designing a T+1 automated distribution system. End-to-end methodology for crystallising operational SOPs into configurable toolkits.', }, }, { tag: { zh: 'AI 独立交付', en: 'Solo AI delivery' }, head: { zh: 'AI 产品独立交付能力', en: 'Ship AI products end-to-end, alone', }, body: { zh: '近一年 solo 设计与开发 AI 对话产品 Axiom（已上线运行，300+ 真实用户、人均对话深度 12 轮），完成从对话工作流架构到 trace 评估的全链路产品决策；在同程主导 AI 智能导览的知识库结构化、B 端后台、chunks 架构设计，从根上解决 AI 讲解的幻觉与多场景复用问题。', en: 'Spent the past year solo-designing and building Axiom, an AI conversation product (live, 300+ real users, 12 turns per session on average) — owning every product decision from dialog workflow architecture through to trace-level evaluation. At Tongcheng, led the knowledge-base structuring, B-side admin tooling, and chunks-level architecture for AI tour guides, solving hallucination at the root while enabling reuse across scenarios.', }, }, { tag: { zh: '业务驱动评估', en: 'Goal-driven evals' }, head: { zh: '业务目标驱动的 AI 产品评估视角', en: 'Business-goal-driven evaluation lens', }, body: { zh: '独立开发开源 LLM 评估框架 PrismEval，并在 Axiom 上建立每日 trace 审查机制，实操对话产品的评估闭环。', en: 'Built and open-sourced PrismEval, an LLM-as-a-judge framework organised by business goal, and runs a daily-trace review loop on Axiom as the manual analogue.', }, }, { tag: { zh: '英文工作', en: 'English work env' }, head: { zh: '全英文工作语境基础', en: 'Comfortable in fully-English work', }, body: { zh: 'Ohayoo 时期对接海外游戏厂商，工作产出（邮件 / 文档 / 需求对齐）以英文为主。', en: 'During Ohayoo I worked with overseas game studios — emails, specs and requirement alignment all in English.', }, }, ], }, // AI projects detail axiom: { name: { zh: 'Axiom · 元溯', en: 'Axiom' }, tagline: { zh: '个人天赋觉察 Agent 应用 · 独立产品', en: 'Personal-talent discovery Agent · Solo product', }, link: 'axiom.nyota.cn', intro: { zh: '独立完成从产品定义、Agent 工作流设计到技术交付的完整链路，验证了从 0 到 1 设计、上线、迭代 AI Agent 产品的端到端能力。上线至今真实用户 300+、人均对话深度 12 轮。', en: 'Solo-shipped end-to-end: product definition → Agent workflow design → technical delivery → iteration. Validates a complete 0→1 capability for AI Agent products. 300+ real users since launch with 12 turns per session on average.', }, decisions: [ { h: { zh: '置信度驱动的非线性 Agent 工作流', en: 'Confidence-driven non-linear Agent workflow', }, b: { zh: '基于「4 基础类型 × 12 亚型」天赋评估理论，设计 6 节点非线性对话工作流（排斥探测 → 心流探测 → 反向验证 → 社交角色 → 副型定位 → 结果合成）。核心设计价值：用置信度驱动的动态路由替代传统线性问卷——AI 根据用户当前响应实时判断置信度，决定下一步走向哪个节点，而不是走固定流程。每节点采用「承接 + 幕后分析 + 兜底选择题」三层 Prompt 结构，在 AI 不确定性下兜底对话推进。这是 AI 对话产品最核心的两个问题——置信度判断与失败兜底——的产品化解法。', en: 'Based on a "4 base × 12 sub-types" talent model, designed a 6-node non-linear dialog workflow (rejection probe → flow probe → reverse-verify → social role → subtype → synthesis). Core design value: confidence-driven dynamic routing replaces a fixed linear questionnaire — the model reads user-response confidence in real time and chooses the next node, rather than following a script. Every node uses a three-layer Prompt scaffold (continuity + behind-the-scenes analysis + fallback multi-choice). This is a productised answer to the two canonical problems of conversational AI: confidence judgement and graceful fallback.', }, }, { h: { zh: '基于真实 trace 的 AI 产品迭代体系', en: 'A trace-grounded iteration system for AI products', }, b: { zh: '上线后建立每日真实对话 trace 审查机制，把模糊的「prompt 调感觉」转化为可追溯的产品质量管理闭环：识别 badcase pattern → 定位具体 prompt 分支 → AB 对比验证 → 上线。所有决策可回溯到一段真实对话——例如某条 trace 显示用户在第 4 轮已表达充分但工作流仍要求其继续展开，定位到对应节点的 prompt 后增加置信度判断分支并 AB 验证，badcase 在下一周的 trace 中显著下降。这套数据驱动的迭代方法论是 Axiom 的迭代引擎，也是 PrismEval 评估框架的实践来源。', en: 'Established a daily real-trace review loop post-launch — turning vague "prompt-by-feel" tuning into a traceable quality-management cycle: spot the badcase pattern → locate the prompt branch → A/B-validate → ship. Every decision traces back to a real conversation — e.g. a trace showed a user already fully expressed by turn 4 while the workflow kept pushing for more; locating the relevant node\'s prompt, I added a confidence-gated branch and A/B-validated it, and the badcase visibly dropped the following week. This data-driven iteration discipline is both Axiom\'s engine and the practical seed for PrismEval.', }, }, { h: { zh: '从用户反馈反推的架构演进决策', en: 'Architecture evolution driven by user feedback', }, b: { zh: 'v1 上线后通过 trace 审查发现两类极端体验——「表达已充分但工作流仍按固定节奏推进」与「表达不足但工作流匆匆结束」。识别根因为固定工作流无法动态判断用户输入密度，决策从扣子（Coze）工作流演进到 Agent 型自建服务，让对话节奏由大模型动态决定。这是从「工作流型对话」到「Agent 型对话」的底层架构升级——也是从被动观察用户反馈到反向定义 AI 架构的一次完整闭环。', en: 'Post-v1 trace review surfaced two failure modes — over-expressed users still dragged through fixed pacing, and under-expressed users hurried to a close. Root cause: fixed workflows can\'t read input density dynamically. Decision: migrate from a Coze workflow to a self-built Agent service — pacing decided by the model itself. A foundational shift from "workflow dialog" to "Agent dialog", and a complete loop from observing user feedback to re-defining the AI architecture.', }, }, { h: { zh: '覆盖完整漏斗的三层商业化架构', en: 'Three-tier monetisation covering the full funnel', }, b: { zh: '免费天赋定位（用户获取）→ ¥9.9 基础测评（首次价值验证 + 付费转化）→ ¥29.9 深度 Agent + 成长追踪（长期留存）。从拉新、价值验证到长期留存的完整商业路径已就位，等待付费数据验证。', en: 'Free talent typing (acquisition) → ¥9.9 basic assessment (first value-validation + paid conversion) → ¥29.9 deep Agent + growth tracking (long-term retention). The full funnel — acquisition → value-validation → retention — is in place, awaiting paid-conversion data.', }, }, { h: { zh: '沉浸式美学交互设计', en: 'Immersive visual & interaction design' }, b: { zh: '「水晶花园」式视觉系统，将 12 种天赋原型可视化为可被用户主动探索的水晶角色集合——把抽象人格类型转化为可被收藏、可被记住的具象 IP，是高沉浸度对话深度（12 轮）背后的体验侧支撑。', en: 'A "Crystal Garden" visual system — 12 talent archetypes rendered as a collectable cast of crystal characters that users can explore actively. Turns abstract personality types into memorable, collectable IPs — the experience-side support behind the 12-turn immersion depth.', }, }, ], tech: { zh: 'v1（已上线）：扣子（Coze）工作流 + Next.js + 阿里云 ECS。完成产品定义与 Agent 工作流设计后，2.5 天独立完成全栈代码交付与上线（与 AI 协作）——具备从产品定义、Agent 工作流设计、技术架构选型到部署上线的完整链路独立运转能力。\nv2（开发中）：自建 Agent 型工作流服务已跑通底层，提供更细粒度 trace 埋点对接 PrismEval。', en: 'v1 (live): Coze workflow + Next.js + Aliyun ECS. After product spec and Agent workflow design were ready, full-stack delivery + launch in 2.5 days, alone (paired with AI). Demonstrates a complete, self-driven loop across product definition, Agent workflow design, tech selection and deployment.\nv2 (in dev): self-built Agent workflow service running, with finer-grained trace instrumentation feeding PrismEval.', }, }, prism: { name: { zh: 'PrismEval', en: 'PrismEval' }, tagline: { zh: '业务目标驱动的 LLM 应用评估框架 · 开源', en: 'Business-goal-driven LLM evaluation framework · Open source', }, link: 'github.com/Nyota-tree/PrismEval', intro: { zh: '现有 AI 评测框架普遍按「技术形式」分类（如 prompt 评测、retrieval 评测、generation 评测），但产品视角需要的是按「业务目标」组织的评估能力。PrismEval 试图填补这一差异。', en: 'Existing evaluation frameworks slice the world by technical layer (prompt eval / retrieval eval / generation eval). The PM lens needs evaluations organised by business goal. PrismEval aims to close that gap.', }, decisions: [ { h: { zh: '设计内核', en: 'Core design' }, b: { zh: '基于 LLM-as-a-judge 范式构建标准化评测链路，核心差异化在于评估指标体系按业务目标驱动——而非按技术分类组织。', en: 'Standardised pipeline on LLM-as-a-judge. The differentiator: a metric system organised by business goal, not by technical category.', }, }, { h: { zh: '当前阶段', en: 'Current stage' }, b: { zh: '完成 prompt 层评测的标准化封装；下一阶段方向是用 Axiom 的真实对话 trace 作为数据集，做 trace 评测的能力补全。', en: 'Prompt-level evaluation is standardised. Next: use Axiom\'s real conversation traces as a dataset and round out trace-level evaluation.', }, }, { h: { zh: '与 Axiom 的协同', en: 'How it pairs with Axiom' }, b: { zh: 'PrismEval 是评估框架的工具化版本，Axiom 的每日 trace 审查是该框架方法论的人工实践版本——两者互为印证与喂养。', en: 'PrismEval is the tooled version of the framework; Axiom\'s daily trace review is the manual version of the same methodology — they validate and feed each other.', }, }, ], }, // Work experience work: [ { org: { zh: '国内 Top3 OTA 平台', en: 'Top-3 OTA Platform in China' }, art: 'ota', role: { zh: 'AI 产品专家', en: 'Senior AI Product Manager' }, period: { zh: '2025.06 – 至今', en: 'Jun 2025 – Present' }, headline: { zh: '主导 AI 对话产品的多场景产品化、知识库架构与 B 端工具包建设', en: 'Owned multi-scenario productisation, KB architecture and B-end toolkit for AI conversational products.', }, summary: { zh: '用 AI 替代景区导览的人力密集型作业模式，2 产品 + 2 研发覆盖 50+ 景区常态化运营，单景区接入周期从行业平均 1 周压缩至 1 天，核心场景 AI 导览渗透率达 70%。', en: 'Replaced labour-intensive in-person tour-guide ops with AI: 2 PMs + 2 engineers covering 50+ scenic spots in steady-state operation; per-spot onboarding compressed from an industry-average week to a single day; AI-tour penetration in core scenarios at 70%.', }, bullets: [ { h: { zh: 'AI 讲解质量诊断与知识库架构重构', en: 'Diagnosing quality issues, rebuilding the KB architecture', }, b: { zh: '通过手工测评定位早期「爬虫 + prompt 改写」方案的核心 badcase pattern——AI 在生成时频繁出现「展品 / 展馆张冠李戴」的事实性幻觉。诊断根因为输入数据形态缺乏结构化事实约束，主导设计「景点级 chunks」方案（含名称、位置、所在场馆、介绍等字段），从数据层强制约束实体归属关系。该方案同时实现三个目标：源头消除幻觉、运营可在后台直接维护字段、一份数据复用讲解 / 知识卡片 / Agent 问答三场景。后续沉淀为团队默认架构，被海南消博会 Agent + RAG 项目直接复用。', en: 'Manual evaluation pinned down the core badcase pattern in the early scrape-then-rewrite pipeline — exhibits and venues being conflated factually. Root cause: unstructured inputs without entity constraints. I led the design of a "spot-level chunks" schema (name / location / venue / description) that enforces entity-belonging at the data layer. It hits three goals at once: hallucination eliminated at source, operations editable directly via admin, single source-of-truth reused across narration / knowledge cards / Agent Q&A. Now the team\'s default; reused by the Hainan Expo Agent + RAG project.', }, }, { h: { zh: 'B 端工具包独立设计', en: 'Solo design of the B-end toolkit' }, b: { zh: '独立设计 AI 导览全套 B 端后台（PRD + 字段定义 + 流程设计），将景区内容运营 SOP 沉淀为可配置工具包，使新景区的内容接入从依赖人工编写转化为标准化数据填充流程。', en: 'Solo-designed the full admin (PRD + field definitions + flows). The scenic-spot ops SOP became a configurable toolkit, turning content onboarding from "write it by hand" into a standard data-filling task.', }, }, { h: { zh: '多场景 AI 产品化扩展', en: 'Multi-scenario productisation' }, b: { zh: '从单一人文景区（恭王府）扩展至自然景区（七星岩、鼎湖山）、动植物园三类场景。基于不同场景下的用户需求差异（人文 - 知识获取、自然 - 出片导航、动植物园 - 知识获取）做差异化产品设计，其中「AI 知识卡片」功能（基于 Markdown 稳定输出 + 优雅版式渲染）为本人独立提出。', en: 'Expanded from cultural sites (Prince Gong Mansion) to nature parks (Qixingyan, Dinghushan) and zoos / botanical gardens. Differentiated product designs per scenario (cultural → knowledge, nature → photo routes, zoo → knowledge). The "AI Knowledge Card" feature (stable Markdown output + elegant typography) was my proposal.', }, }, { h: { zh: '从 0 到 1 接入 Agent + RAG 能力 — 海南消博会', en: '0→1 Agent + RAG at the Hainan Expo', }, b: { zh: '在政府级展会场景下从 0 引入 AI Agent + RAG 架构，主导意图识别的场景定义与 chunks 语料结构对齐，完成对话产品的端到端落地。', en: 'Introduced Agent + RAG from scratch into a government-level expo. Led intent-recognition scenario definition and corpus-structure alignment, delivering the dialog product end to end.', }, }, { h: { zh: 'B 端商业模式设计', en: 'B-end business-model design' }, b: { zh: '构建「AI 工具免费铺设渗透 → 换取售票权」的 B 端打法，借助景区「AI 转型政绩诉求」与「用户免费体验诉求」双轮驱动，跑通 AI 产品在文旅景区的商业化路径。', en: 'Designed a "free AI tooling for ticketing rights" play, riding the dual demand for AI-transformation credentials on the operator side and free experience on the user side. Validated a commercialisation path for AI products in cultural tourism.', }, }, ], }, { org: { zh: '国内 Top 金融互联网公司', en: 'Top Fintech Internet Company in China' }, art: 'fintech', role: { zh: '产品运营专家', en: 'Senior Product Operations' }, period: { zh: '2024.06 – 2025.06', en: 'Jun 2024 – Jun 2025' }, headline: { zh: '接手 10 年生命周期的成熟产品，在维稳基调下推动用户增长创新', en: 'Took over a 10-year-old mature product and pushed growth innovation under a stability mandate.', }, bullets: [ { h: { zh: '关键留存断崖点的用户行为洞察', en: 'Identifying the retention cliff', }, b: { zh: '通过用户行为数据分析定位芭芭农场数值体系的核心断崖点——大量用户停留在中间级别放弃（受限于游戏后期需要超过生命周期 50% 数值才能升到满级的设计），沉睡用户的累积数值资产在原机制下完全浪费。', en: 'Behavioural analysis pinpointed the core cliff in Baba Farm\'s progression: a large mass of users abandoned at mid-level (the late game requires >50% lifetime value to max out). Dormant users\' accumulated assets were effectively wasted under the old mechanic.', }, }, { h: { zh: '「亲友帮帮种」双边激励机制设计', en: '"Friends-help-grow" two-sided incentive mechanic', }, b: { zh: '· 设计逻辑：高级别用户（需要数值加速）+ 沉睡中级用户（数值资产被浪费）的不对称需求匹配；\n· 机制核心：沉睡用户的「已放弃树」灌注给活跃高级别用户作为助力，最高获得 5 倍数值加速；\n· 社交关系长期化：助力关系沉淀为可复用「队伍」，支持后续共种，让单次激励演化为长期社交粘性；\n· 业务成果：通过 5% 留存桶 A/B 测试验证，DAU 增量正向显著，成为芭芭农场 2024 年度最大增长引擎。', en: '· Design logic: an asymmetric match between high-level users (needing acceleration) and dormant mid-level users (with wasted assets).\n· Mechanic: dormant users\' "abandoned trees" channel into active high-level users as boosts — up to 5× acceleration.\n· Long-term social: helper relationships persist as reusable "teams" supporting later co-growing — single-shot incentive becomes long-term social glue.\n· Results: validated by a 5% retention-bucket A/B; significant positive DAU lift. The biggest growth engine of Baba Farm in 2024.', }, }, { h: { zh: '跨集团专项推动', en: 'Cross-group programme delivery' }, b: { zh: '芭芭农场作为蚂蚁 + 淘天双集团联合产品（入口覆盖支付宝、淘宝、饿了么），数据完全互通。本人作为项目负责人主导该项目跨集团推进。', en: 'Baba Farm sits across Ant + Taobao groups (entrances on Alipay, Taobao, Ele.me; data fully shared). As project lead, I drove cross-group delivery end to end.', }, }, ], }, { org: { zh: '国内 Top1 直播平台', en: 'China\'s #1 Live-streaming Platform' }, art: 'live', role: { zh: '产品运营专家', en: 'Senior Product Operations' }, period: { zh: '2021.07 – 2024.06', en: 'Jul 2021 – Jun 2024' }, headline: { zh: '开放平台 C 端运营 → 主播工具开放平台 → 弹幕玩法主播侧产品运营', en: 'Open-platform C-end ops → anchor-tool open platform → bullet-screen mini-game anchor-side ops.', }, summary: { zh: '三阶段业务转型期间持续在双边开放平台场景中独立交付核心机制设计。', en: 'Across three business pivots, kept solo-shipping core mechanism designs inside two-sided open-platform contexts.', }, phases: [ { title: { zh: '第一阶段：概率玩法开放平台 C 端运营 (2021.07 – 2023.04)', en: 'Phase I — C-end ops for the probability-game open platform (Jul 2021 – Apr 2023)', }, intro: { zh: '本人作为 C 端运营 owner 负责机制设计与规则管控。期间业务规模从 0-1 高速增长，在大业务体系内承担必不可少的份额。', en: 'I owned C-end ops — mechanic design and rule governance. The business grew rapidly from 0 to 1, holding an essential share of revenue inside the broader business unit.', }, items: [ { h: { zh: '付费档位重设计 — 数据驱动的两轮迭代', en: 'Payment-tier redesign — two data-driven iterations', }, b: { zh: '通过付费曲线分析两次定位拐点，先后增加 100 元档与 10 元档付费选项，通过 A/B 测试验证两次累计带来 67% 流水增量。', en: 'Pay-curve analysis surfaced inflection points twice. Adding the ¥100 and later the ¥10 tier (each A/B-tested) drove a cumulative +67% in revenue.', }, }, { h: { zh: '基于 UGC 传播路径的活动机制设计', en: 'Campaign mechanic built on UGC propagation', }, b: { zh: '识别「主播获得高价值礼物 → 主动发布短视频」的核心传播链路，主导推动周期性礼物轮换机制；针对传播爽点设计「低概率高奖励」结构（区别于均匀提升概率），活动流水提升 15%。', en: 'Spotted the core loop of "anchor receives a high-value gift → posts a short video" and drove a periodic gift-rotation mechanic. A low-probability / high-reward structure (vs. uniform odds) tuned for the share-worthy moment; +15% campaign revenue.', }, }, { h: { zh: '用户分层与产品暴露策略', en: 'Tiered user segmentation and exposure', }, b: { zh: '基于 ARPU 设计差异化暴露——高付费用户暴露高刺激高客单玩法，低付费用户暴露轻量化低风险玩法。', en: 'ARPU-based differentiated exposure — high payers see higher-stimulus, higher-ticket games; low payers see lighter, lower-risk ones.', }, }, { h: { zh: '玩法评级与流量分配机制', en: 'Game-rating and traffic-allocation system', }, b: { zh: '建立三维评级体系（舆情次数 + bug 次数 + 流水绝对值），配套自动化触达控制——对流水过热玩法主动屏蔽未触达用户与近一月未付费用户，避免负向循环。', en: 'Three-axis rating (PR incidents + bug count + absolute revenue) with automated reach controls — overheating games are masked from un-reached users and users who haven\'t paid in the last month, preventing negative loops.', }, }, { h: { zh: '早期信号识别 + 主动降速', en: 'Early-signal detection + intentional slowdown', }, b: { zh: '针对概率玩法「一旦付费即高粘性」特征，设计实时监控 + 跨产品基线对比，在流水进入危险区间主动长期管控新增。核心产品判断：主动牺牲短期增长以保护长期生态。', en: 'Probability games have "pay-once-then-stick" characteristics. Built real-time monitoring + cross-product baseline comparison; once revenue entered the danger zone, new-user acquisition was throttled long-term. Core product call: trade short-term growth for long-term ecosystem health.', }, }, { h: { zh: '高风险业务的合规管控体系', en: 'Compliance governance for high-risk business', }, b: { zh: '独立设计业务侧安全审核标准；建立用户付费基准 + 限额 + 防沉迷机制；设计历史付费回查与异常退款追回流程；与内容安全团队协作建立直播间洗钱与异常流量的长期监控。', en: 'Designed business-side safety review standards; built payment baselines + caps + anti-addiction; designed retrospective-payment audit and anomalous-refund recovery flows; partnered with content-safety on long-term monitoring for live-stream money-laundering and traffic anomalies.', }, }, ], }, { title: { zh: '第二阶段：主播挂载工具开放平台 (2023.04 – 2023.Q3)', en: 'Phase II — Anchor-tool open platform (Apr 2023 – Q3 2023)', }, intro: { zh: '团队战略转型期间，独立从 0 到 1 建设主播挂载工具开放平台。', en: 'During the team\'s strategic pivot, solo-built the anchor-tool open platform from 0 to 1.', }, items: [ { h: { zh: '从 0 到 1 独立建设', en: 'Solo 0→1 build' }, b: { zh: '作为唯一负责人完成产品设计、运营、跨前后端需求对接。', en: 'Sole owner across product design, ops and front/back-end requirement alignment.', }, }, { h: { zh: '业务洞察 — 解决长尾合规问题', en: 'Insight — solving long-tail compliance', }, b: { zh: '观察到主播侧存在「野生工具」（解决长尾差异化需求但爬取用户信息）的合规风险，平台官方 PK 工具的通用性又满足不了长尾需求。通过开放平台让开发者侧供给长尾能力，同时把数据安全收敛回平台。', en: 'Anchors were using off-platform "wild" tools that met long-tail needs but scraped user data — a compliance risk — while official PK tooling was too generic. The open platform lets developers serve long-tail demand while pulling data security back into the platform.', }, }, { h: { zh: '首创直播业务「按月订阅」商业模式', en: 'First "monthly subscription" business model in live', }, b: { zh: '在传统直播打赏 / 单次付费之外，独立设计开放平台开发者侧的按月订阅模式——开发者按月付费获得平台坑位与能力调用，平台获得稳定可预测的现金流，同时绑定开发者长期投入。这是直播行业首个面向开发者的订阅化商业模式落地。', en: 'Beyond traditional tipping / one-off purchases, designed a developer-side monthly subscription for the open platform — developers pay monthly for slots and capability calls; the platform gains predictable cash flow and binds developers to long-term investment. The first developer-facing subscription model shipped in the live-streaming industry.', }, }, ], }, { title: { zh: '第三阶段：弹幕玩法 — 主播侧产品与运营 (2023.Q3 – 2024.06)', en: 'Phase III — Bullet-screen games, anchor-side product & ops (Q3 2023 – Jun 2024)', }, intro: { zh: '全权负责开放平台主播资源侧工作（虚线带 1 人）。', en: 'Owned the anchor-resources side of the open platform end to end (dotted-line lead of 1).', }, items: [ { h: { zh: '运营 SOP 产品化 — 从人工配置到 T+1 自动化分发系统', en: 'Productising the SOP — from manual config to a T+1 automated distribution system', }, b: { zh: '诊断推荐位「运营每周人工配置」机制的三大缺陷（不敏捷 / 新玩法无法识别 / 不可规模化），主导设计三层自动化分发架构——\n· 坑位分层：新游坑位（保护新供给）+ 老游坑位（维持精品流水稳定）；\n· 新游预评级机制：上线 7 天内进入预评级，达标自动进入推荐位；\n· 老游异动识别 + 绿色通道：流水异动超阈值进入扶持通道，稳定后自动升级评级；\n· 底层为日级评级模型 + T+1 推荐位更新，将运营 SOP 完全机制化、自动化。', en: 'Diagnosed three flaws in the weekly-manual-config recommendation slot (slow, blind to new games, non-scalable). Led a three-layer automated distribution architecture:\n· Slot tiering — new-game slots (protecting new supply) + legacy slots (stable revenue from polished games).\n· New-game pre-rating — entering pre-rating within 7 days of launch; passing the bar promotes them into recommendation slots automatically.\n· Legacy anomaly detection + fast lane — revenue swings beyond a threshold enter a support lane and auto-upgrade once stable.\n· Underneath: a daily rating model + T+1 slot updates — fully mechanising and automating the ops SOP.', }, }, { h: { zh: '跨业务能力迁移 — 玩法发行画像方法论', en: 'Cross-business transfer — game-publishing profile methodology', }, b: { zh: '基于 Ohayoo 游戏发行经验，从历史数据反推「高潜玩法画像」，沉淀为开发者培训营内容 + 主播侧推荐策略，从培训营定向跑出 2 款 Top3 级别玩法，占整个平台流水 10%。', en: 'Reverse-engineered a "high-potential game profile" from Ohayoo publishing history, then turned it into developer-bootcamp content + anchor-side recommendation strategy. The bootcamp produced two Top-3-tier games — 10% of the platform\'s revenue.', }, }, { h: { zh: '冷启策略迭代设计', en: 'Cold-start strategy redesign' }, b: { zh: '识别概率玩法时代（主播主动挂载）与弹幕玩法时代（营收价值不明确）的差异，设计「内测群 → 内测数据 → 流量分配」的三段式冷启机制。', en: 'Spotted the gap between the probability-game era (anchors actively mounted tools) and the bullet-screen era (unclear monetisation). Designed a three-stage cold start: closed-beta group → closed-beta data → traffic allocation.', }, }, { h: { zh: '主播侧推荐策略与运营', en: 'Anchor-side recommendation and ops', }, b: { zh: '通过推荐策略优化与主播侧运营，直播伴侣端日均新增 +230%、整体挂载 UV 提升 50%、主播渗透率达 15%。', en: 'Recommendation tuning + anchor-side ops drove +230% in daily new on the Live Companion app, +50% mount UV overall, and 15% anchor penetration.', }, }, ], }, ], }, { org: { zh: 'Ohayoo', en: 'Ohayoo' }, art: 'game', role: { zh: '游戏发行', en: 'Game Publishing' }, period: { zh: '2020.09 – 2021.07', en: 'Sep 2020 – Jul 2021' }, summary: { zh: '负责自研及海外引入休闲游戏的发行、核心玩法调优与商业化落地。对接海外游戏厂商，工作产出（邮件 / 文档 / 需求对齐）以英文为主。', en: 'Owned publishing, gameplay tuning and monetisation for in-house and overseas casual titles. Worked with overseas studios — emails, specs, alignment all in English.', }, }, { org: { zh: '雷霆游戏', en: 'G-bits / Thunder Games' }, role: { zh: '游戏发行', en: 'Game Publishing' }, period: { zh: '2019.07 – 2020.09', en: 'Jul 2019 – Sep 2020' }, summary: { zh: '负责 10+ 款精品微信小游戏及重磅 IP 手游《摩尔庄园》的发行工作，积累泛娱乐用户洞察与游戏化交互设计经验。', en: 'Published 10+ premium WeChat mini-games and the flagship mobile IP Mole\'s World — built up entertainment-user insight and gamification design fundamentals.', }, }, { org: { zh: '华南理工大学', en: 'South China University of Technology' }, role: { zh: '信息与交互设计 · 本科', en: 'BA, Information & Interaction Design' }, period: { zh: '2015.09 – 2019.06', en: 'Sep 2015 – Jun 2019' }, summary: { zh: '信息与交互设计专业本科。', en: 'Bachelor of Arts in Information & Interaction Design.', }, }, ], awards: { title: { zh: '奖项与影响力', en: 'Recognition & community' }, bytedance: { title: { zh: '字节跳动内部认可', en: 'Internal recognition at ByteDance' }, items: [ { zh: '字节范 (2023 Q2)', en: 'ByteStyle Award (2023 Q2)' }, { zh: '直播部门业绩之星 (2022.07–08)', en: 'Live Dept. Performance Star (Jul–Aug 2022)', }, { zh: 'Ohayoo 最有字节范奖 (2021.01–02)', en: 'Ohayoo Most-ByteStyle (Jan–Feb 2021)', }, { zh: '字节跳动公司内部 ERG 女性员工资源小组负责人', en: 'Lead, Internal Women\'s ERG at ByteDance', }, ], }, community: { title: { zh: '社区与领导力', en: 'Community & leadership' }, items: [ { zh: '创业森林黑客松 二等奖 队长', en: 'Startup Forest Hackathon — 2nd prize, captain' }, { zh: '个人公众号「Nyota 佳树」，运营 3 个月粉丝达到 1000+', en: 'Personal column "Nyota 佳树" — grew to 1k+ in 3 months', }, { zh: '《复业训练营》合伙人（负责体验设计，首期触达 250+ 学员）', en: 'Partner at Side-Career Bootcamp — owned experience design, 250+ first-cohort students', }, ], }, }, }; window.RESUME = RESUME;