兼具商业洞察与技术认知的产品经理——专注 AI Agent 系统、AIGC 内容链路、流量智能投放三大方向，善于从数据中找策略、从策略中跑增长。 Product Manager bridging business strategy and engineering fluency — specializing in AI Agent systems, AIGC content pipelines, and intelligent traffic delivery. Turning data into strategy, strategy into measurable growth.
核心能力What I Bring
数据驱动 × AI 工程认知 × 流量商业化全链路 Data-driven thinking × AI engineering fluency × Full-funnel traffic monetization
0-1 搭建盯盘、诊断、复盘、问答四大智能体，深度应用 Prompt Engineering、RAG、多轮对话设计，日均服务 500+ 咨询。Built 0-to-1 Agent skills for monitoring, diagnosis, post-review & Q&A. Deep Prompt Engineering, RAG, and multi-turn dialogue expertise. 500+ daily queries.
设计流量分发、用户控量、赛马机制全链路策略，平台时长负向改善 43%+，涨粉率提升 73%，日均价值增量 20 万元。Designed distribution throttling, user-level control, and A/B creative strategy. Dwell-time improved 43%+, fan growth +73%, ¥200K daily incremental value.
主导多模态文案与图片生成平台，日均调用 1300 次，CTR +23%，节约设计 7 人/天，订单 AI 渗透率 +24%。Led multimodal copywriting & image generation platform. 1,300 daily calls, +23% CTR, saves 7 designer-days/day, +24% AI order penetration.
搭建素材全链路洞察工具（投前选品→投中监控→投后画像），打通投放与用户数据，投中监控效率 +60%。Built full-funnel creative insight tools (selection → live monitoring → post-campaign profiling). Connected delivery & user data; +60% monitoring efficiency.
CUHK 信息系统硕士（QS前50），IELTS 7分，滴滴国际化外卖数据分析经验，熟悉港、新商业环境。MSc from CUHK (QS Top 50), IELTS 7.0. DiDi international food delivery analytics. Comfortable in HK/SG business environments.
掌握 LLM/RAG/SFT 基础流程、LangChain Agent 框架、Java & Python、SQL(Hive)，能有效跨越产研协作边界。Fluent in LLM/RAG/SFT pipelines, LangChain Agents, Java & Python, SQL (Hive). Bridges product and engineering teams effectively.
项目经历Work Experience
点击项目卡片展开详细成果Click cards to expand project details
AI Native 投放平台，落地盯盘、诊断、复盘、问答四大 Skill，0-1 完整闭环。Built AI-native delivery platform with 4 Agent skills: monitoring, diagnosis, post-review, Q&A — complete 0-to-1 cycle.
- 平台搭建：对标 10+ 平台，搭建 AI Native 投放平台，集成外部依赖平台技能，上线操作任务定时自动化执行能力。Benchmarked 10+ platforms; built AI-native delivery platform with automated scheduled task execution.
- 盯盘 Skill：梳理投中异常识别场景，结合 Prompt 工程优化盯盘逻辑，梳理内外部数据依赖，0-1 上线盯盘 Skill。Mapped anomaly detection scenarios, optimized monitoring logic with Prompt Engineering, launched monitoring Skill from 0-to-1.
- 诊断能力：打磨投放计划和内容维度 30+ 诊断项，白盒化引擎数据，功能用户渗透率 30%。Refined 30+ diagnostic dimensions across delivery plans and content; whitebox engine data; 30% user penetration.
- 复盘能力：收集 20+ 报告样例，抽象五组场景，支持场景化复盘和报告自动执行，预计节约 7 人日/次。Collected 20+ report samples, abstracted 5 scenario types, automated report generation saving 7 person-days.
- 投放答疑：整合规则库与历史案例，实现投前常见问题实时响应，准确率 82%，日均解决 500+ 咨询，减少人工 40%。Integrated rules & case library; 82% accuracy, 500+ daily queries, 40% reduction in manual workload.
多模态、多版位投放素材生成能力，解决素材质量波动大、设计人力紧张的核心痛点。Multimodal, multi-placement creative generation pipeline; solved quality inconsistency and design resource scarcity.
- 素材生产：上线文案和图片生成功能，日均调用 1300 次，CTR 提升 23%，节约设计人力 7 人/天。Launched copywriting & image generation; 1,300 daily calls, +23% CTR, saves 7 designer-days/day.
- 策略迭代：定期沉淀优质文案迭代模型，采纳率提升 15%；自动填充能力使订单 AI 渗透率提升 24%。Continually fine-tuned model with high-quality samples, +15% adoption; auto-fill drove +24% AI order penetration.
- 赛马投放：优化多版本素材赛马投放，策略渗透率 17%，累计 CTR 提升 17%。Optimized multi-variant creative racing; 17% strategy penetration, cumulative +17% CTR.
投放全链路素材洞察（投前选品→投中监控→投后画像），解决创作无参考、调优慢、复盘无方向的问题。Full-funnel creative insight from pre-launch selection to live monitoring and post-campaign profiling.
- 投前选品：跑量素材筛选模型 + AI 总结垂类规律，预算消耗率 +8%，涨粉率 +8%。Pre-launch: high-performing creative filter + AI vertical pattern summary; +8% budget efficiency, +8% fan growth.
- 投中大盘：专属运营看板，整合在投数量、过审效率、标签分布，一键清理低质素材，效率 +60%。Live monitoring dashboard with approval rates, tag distribution, one-click low-quality cleanup; +60% efficiency.
- 投后画像：打通投放与用户数据，满足多链路、多位置的用户标签诉求。Connected delivery and user data for multi-funnel user profiling needs.
流量分发、筛选及分配策略设计，解决平台时长负向超标与投放效率偏低的核心问题。Traffic distribution, filtering and allocation strategy to fix over-indexing on high-value users and low delivery efficiency.
- 诊断：识别三大问题——高价值用户过度集中、分发规则导致匹配度低、固定权重推送非感兴趣内容。Identified 3 root causes: over-concentration on high-value users, low content-user match, fixed-weight misalignment.
- 策略：流量+用户双维度控量，平衡高峰低谷，前置筛选低价值流量，动态调整分配权重。Dual-dimension throttling (traffic + user), smoothed peak-valley imbalance, pre-filtered low-value impressions, dynamic weight adjustment.
- 效果：主站时长负向优化 43.61%，极速版 53.64%，长播率 +68.06%，涨粉率 +73.12%，日均流量价值增量 20 万元。Main app dwell-time negative rate −43.61%, Lite −53.64%; long-play rate +68.06%, fan growth +73.12%; ¥200K daily incremental value.
技能体系Skills & Expertise
教育背景Education
联系我Let's Connect
欢迎潜在雇主及合作方联系 · 点击卡片可自定义链接 Open to employers & collaborators · Click cards to customize links