Welcome to my corner of the web

London / Beijing

Hey, I'm
Zhao Zhengyao!

Dual degree in Communication and Computer Science—I bring both critical thinking from the social sciences and engineering rigor. Now focused on AI and AI products, and a heavy user of AI writing and AI coding tools.

Previously incubated a 0-to-1 AI co-creation community at Meituan and drove commercialization growth at Tencent QQ.

LSE MSc Media & Communications (Data & Society)HUST Communication & Computer Science Dual DegreeGPA 89.9 & 90.3

Scroll down to see what I've been up to.

Send Email
GitHubLinkedInXiaohongshu
1430439311@qq.com15731698896
Profile photo

Projects

End-to-end execution from research to deployment

YangMenu

YangMenu

Try it online →

WeChat mini-program helping Chinese students and travelers understand overseas restaurant menus

Role: Product Design · Full-Stack Development

Feb 2026 - Present

TypeScriptVant WeappWeChat CloudBaseDeepSeek-VL2

Photo menu recognition → AI translation, dish explanation, flavor notes, allergen alerts. End-to-end solo build from insight to deployment; developed with Vibe Coding (Cursor + Claude Code).

  • Core flow: photo → AI translation, dish explanation, flavor notes, allergen alerts
  • End-to-end solo: insight, tech selection, development, launch
  • Vibe Coding workflow: Cursor + Claude Code for efficient development
Xiaohongshu AI Content Generator

Xiaohongshu AI Content Generator

A custom AI writing system for "你的AI背包", solving hallucination and style drift in generic AI tools

Role: Product Design · Prompt Architecture · Content Operations

Feb 2026 - Present

Prompt EngineeringContent OperationsSEO Optimization

While running a Xiaohongshu account, I found generic AI content suffered from three problems: fabricated personal anecdotes, inconsistent tone, and ignored platform rules. Core approach: embed quality standards into the Prompt system so AI aligns at generation time, not through post-hoc editing.

  • Three-layer architecture: Anti-hallucination layer (user supplies context each time, eliminating fabrication), Style control layer (four content structure templates), Platform layer (title length limits, SEO keyword placement, mobile-friendly formatting)
  • Iterated from v1.0 to v2.0: data analysis of 5 posts revealed search keywords as the key traffic driver, leading to an SEO module and a 6-item post-generation checklist
  • Account built from zero — third post achieved 1,600+ saves and 1,000+ likes
AI Academic Writing Assistant for Social Sciences

AI Academic Writing Assistant for Social Sciences

Try it online →
In development

RAG-based writing tool with private literature library, addressing citation hallucination

Role: Solo design and development

RAGVector retrievalLLM API

Three-panel UI (library—AI Q&A—editor). Batch literature selection and AI-guided writing; precise citation from library, enabling deep collaboration where humans handle argumentation and AI handles text execution.

  • Born from real needs in social science writing; maintains rigor while boosting efficiency
  • Three-panel design: library, AI Q&A, editor in sync
  • Precise citation: humans for logic, AI for text execution
Federated Boosting with Differential Privacy

Federated Boosting with Differential Privacy

HUST undergraduate thesis — collaborative model training without centralizing user data, balancing performance and privacy

Role: Core Researcher

Jun 2024

PyTorchCUDARayTensorBoardNumPy

Built on Shen et al. (2022)'s Federated Functional Gradient Boosting (FFGB) algorithm, extending it with a privacy protection mechanism. Clients train weak learners via residual tracking each round; the server aggregates them into a global model. An adaptive Gaussian noise mechanism dynamically adjusts noise intensity based on loss changes to balance privacy and accuracy.

  • Integrated Boosting into the federated framework with an adaptive Gaussian differential privacy mechanism that tunes noise strength dynamically
  • Applied knowledge distillation to compress the ensemble model and reduce client–server communication overhead
  • Validated across multiple datasets under high data heterogeneity: significantly faster convergence than FedAvg, with strong model accuracy retained after applying differential privacy

Experience

Meituan

Meituan

Core Local Commerce | AI Product Manager Intern

May 2025 - Aug 2025Beijing

Incubated the "AI Co-creation Community" from 0-to-1 to address high AI adoption barriers and repetitive work in product and ops teams, leveraging AI-assisted full-stack development. The project aimed to establish a cross-functional collaborative development model and build an AI knowledge base to elevate org-wide AI capability.

  • -User Research & Product Design: Surveyed 50+ product and ops staff, identified 8 high-frequency scenarios (e.g., daily reports, meeting summaries), and filtered for use cases achievable by non-engineers with AI—no dev sprint needed. Led three-module product design (Project Incubation, Knowledge Sharing, Tool Integration), delivering state diagrams and PRD with complete interaction logic.
  • -AI-Assisted Full-Stack Development: Using AI coding tools (Cursor et al.), the product team independently completed front-end, back-end, and database deployment—bypassing traditional review and sprint cycles—compressing the MVP launch timeline from months to 20 days. Co-authored collaboration rules (tech stack constraints, Git commit standards) and ensured output quality through AI cross-review.
  • -Cold Start & Scaled Operations: Simultaneously launched across 4 internal groups of 500 members each, amplified by company-wide promotion. Within two months, reached 2,000+ employees across 5 business lines, yielding 41 AI efficiency projects and reusable components (SSO integration, LLM API access). The Smart Daily Report Assistant—a standout project—supports user-customized Prompt configuration and saw widespread adoption after launch.
Tencent

Tencent

QQ | Product Manager Summer Intern

Apr 2024 - Jan 2025Shenzhen

Led product planning and operations for Yuanstore (次元市集), QQ’s IP merchandise marketplace for anime collectibles such as trading cards and pins.

  • -B2B Supply Chain Overhaul: Redesigned the merchandising system to fix slow product listing. Built a batch upload feature and standardized SOPs, doubling listing efficiency (from 3 to 6 SKU sets/week, +100%) and directly driving weekly GMV up 25%.
  • -Merchandising & Catalog Expansion: Combined QQ in-app user data with external trend signals (e.g., Xiaohongshu) to design an IP heat evaluation framework for targeted high-potential IP discovery. Expanded the direct-supply catalog 4x through procurement and B2B partnerships, resolving the platform’s core SKU gap.
  • -Commercialization Campaigns: Built standardized campaign configuration templates (H5 templates & promo rules) to reduce ops overhead. Led multiple themed promotional campaigns, growing monthly GMV from 469K to 811K (+73%).
  • -Customer Service System Upgrade: Designed a CS monitoring dashboard and alert system. Set KPIs including "30s response rate"; improved response rate from 40% to 90% and achieved zero complaints for the quarter.

Education

London School of Economics

London School of Economics

MSc Media and Communications (Data & Society)

Sep 2025 - Dec 2026
London, UKGPA: In Progress
  • -MSc Media and Communications Programme
Data & SocietyMedia IndustriesAlgorithmic GovernanceComputational Communication
Huazhong University of Sci. & Tech.

Huazhong University of Sci. & Tech.

Communication (Network Communication) & Computer Science Dual Degree

Sep 2021 - Jun 2025
Wuhan, ChinaGPA: GPA 89.9 & 90.3
  • -Outstanding Intern (Top 4%)
  • -Communication & Computer Science Dual Degree
App Innovation Design (98)Data Mining (97)Algorithm Design (90)

Skills

Languages

English (Business)IELTS 7.0PTE 81/90CET-6 500+Mandarin Level 1 Grade B

Programming & Software

FigmaMockplusExcel (Expert)PythonC++XmindCapCut