Andrew Yao logo

AI PM · BUILDER · 2026

PORTFOLIO / 2026Chengdu, China

AI Product Manager / Independent Builder

I turn ambiguous product ideas into AI products that users can use, teams can evaluate, and engineers can ship.

SELECTED SYSTEMS

  1. 01AI GTM SYSTEM
  2. 02ORMA
  3. 03BUZZY
  4. 04SIDER
Open to AI product roles and thoughtful collaborationsSCROLL TO READ
00POINT OF VIEW / PRODUCT THINKING

AI MAKES SHIPPING CHEAPER.

Evidence becomes the bottleneck.

Pages, ads, and prototypes can all be generated faster. The hard questions remain: who cares, which message holds up, and what signal deserves more investment?

My work is to translate across users, product, technology, and business—turning model capabilities into understandable experiences and market feedback into decisions a team can act on.

01SELECTED WORK

Systems, not screens.

  1. 01AI GTM SYSTEMSTEALTH AI PRODUCT · 0→12026 — NOW
  2. 02ORMABRAND-TO-CAMPAIGN WORKSPACE2026 — NOW
  3. 03BUZZYAI ADVERTISING WORKFLOW2025.08 — 2025.12
  4. 04SIDERWEB CREATOR & WISEBASE2024.09 — 2025.08
01STEALTH AI PRODUCT · 0→12026 — NOW

AI GTM SYSTEM

A system that turns product facts into testable GTM hypotheses, real market touchpoints, and growth signals—helping early teams decide what to do before writing more code or buying traffic.

THE QUESTION

AI has made pages, copy, and ads inexpensive to produce. Early teams still lack a continuous workspace that connects hypotheses, launches, signals, and decisions.

WHAT I SHAPED

  • Defined the product loop from Brand → GTM Loop → Surface → Signal → Decision Brief
  • Designed multi-agent collaboration, workspace canvases, landing-page generation, and publishing flows
  • Worked across positioning, pricing, SEO, Meta Ads synchronization, and technical architecture
  • Used the real building process to test product boundaries instead of optimizing for an AI demo
  • Product Strategy
  • Agent Workflow
  • GTM Experiment
  • Full-stack
02BRAND-TO-CAMPAIGN WORKSPACE2026 — NOW

ORMA

A creative workspace built around Brand → Campaign → Canvas, keeping brand facts, audiences, style, and channel content connected through one context chain.

THE QUESTION

Marketing context is often lost between tools and roles, causing brand facts, campaign goals, and final channel assets to drift apart.

WHAT I SHAPED

  • Decomposed complex campaigns into seven bounded, orchestratable canvas nodes
  • Made channel content a natural output of the strategy chain instead of an isolated generation result
  • Defined a complete product structure across Web, API, agents, prompt management, and telemetry
  • Canvas Product
  • AI Orchestration
  • Prompt System
  • Product Architecture
03AI ADVERTISING WORKFLOW2025.08 — 2025.12

BUZZY

An AI workflow for small advertising teams and creators that connects competitive research, campaign planning, and video generation.

THE QUESTION

Creative production takes time, competitor research depends on manual work, and planning is disconnected from video generation. Small teams struggle to sustain the cost of iteration.

WHAT I SHAPED

  • Led more than 15 user interviews and systematically compared eight competing products
  • Defined a differentiated research → planning → generation journey and a free-form canvas
  • Designed four subscription tiers and a credit model to balance access, compute cost, and gross margin
  • Designed a multi-agent architecture using YAML orchestration, Temporal execution, and multimodal models
04WEB CREATOR & WISEBASE2024.09 — 2025.08

SIDER

Across an AI no-code site builder and a personal knowledge base, I translated between product definition, RAG design, prompts, backend delivery, and growth.

THE QUESTION

Enable non-technical users to move from description to generation and refinement, while making file management, semantic search, and AI conversations a dependable knowledge workflow.

WHAT I SHAPED

  • Studied Bolt.new, Lovable, V0, and Figma-to-code tools to define the non-technical audience
  • Shaped core interactions and prompts, and delivered key WebCreator backend services
  • Defined Wisebase file management, RAG semantic search, and AI conversation capabilities
  • Designed document chunking, vectorization, retrieval strategy, and the foundational service framework
02THE INDEPENDENT CHAPTER
2025.12 — NOWNO GAP. A DIFFERENT KIND OF WORK.

Independent AI Product Builder

This is not a career gap. It is a founder chapter in which I work hands-on from problem selection, positioning, prototyping, and system design through code, deployment, and market narrative.

AI makes shipping cheaper. Product judgment—and evidence—become the bottleneck.
  1. 01RESEARCH
  2. 02FRAME
  3. 03PROTOTYPE
  4. 04BUILD
  5. 05LAUNCH
  6. 06LEARN
03EXPERIENCE

Engineer by training. Product by practice.

From edge devices and infrastructure to AI workflows and commercial products. Technology is not a separate path; it is how I understand constraints and move product work into delivery.

  1. 2025.08 — 2025.12
    ZMO
    Technical Product Manager

    Led requirements, product design, technical reviews, monetization design, and cross-team delivery for an AI video creation product line.

  2. 2024.09 — 2025.08
    SiderWeb
    Technical Product Manager

    Owned market and user analysis, requirements, and technical solution design while contributing to backend, prompt, agent, and RAG implementation.

  3. 2023.11 — 2024.09
    自研 AI 笔记
    Product Lead / Builder

    Took an independent AI notes product from market research and Figma prototypes through full-stack development, AI features, subscription payments, cloud deployment, and operation.

  4. 2021.09 — 2023.11
    科大讯飞AI Learning
    Software Engineer

    Built infrastructure, a task-flow framework, OTA capabilities, device services, and internal open-source foundations for an AI translation pen.

  5. 2021.04 — 2021.08
    罗克佳华集团IoT Platform
    Software Engineer

    Contributed to IoT platform positioning and architecture, including multi-protocol device access, device shadows, and business automation.

  6. 2019.06 — 2021.04
    成都鹏业软件Research Institute
    Software Engineer

    Led feature design, hardware selection, and the video-stream framework for a Jetson edge-recognition terminal, and contributed to the internal IoT platform.

04WORKING RANGE

From ambiguity to signal.

A

PRODUCT JUDGMENT

Find the problem worth solving before deciding what to generate.

  • User Research
  • Positioning
  • Competitive Analysis
  • Pricing
  • Roadmap
B

AI SYSTEMS

Organize model capabilities into workflows users understand and teams can maintain.

  • Agent Workflow
  • RAG
  • Prompt Design
  • Evaluation
  • Orchestration
C

SHIP

Move from product definition to software that runs, deploys, and improves.

  • React
  • TypeScript
  • Go
  • Python
  • Cloud & Infrastructure
D

GROWTH

Bring market signals back into product decisions instead of stopping at traffic reports.

  • GTM Experiment
  • SEO
  • Meta Ads
  • Analytics
  • Decision Brief

OUTSIDE THE WORKFLOW

I also photograph people and streets.

Photography reminds me that real people do not follow flowcharts. Light, distance, hesitation, and emotion often reveal whether an experience works before the metrics do.

06CONTACT

Let's build something people can understand—and use.

Email Andrew