It remembers your context and guides next actions across applications and interviews.
View case studyFeiya SHU — AI Product Manager
I research,
build
and ship
human–AI
systems.
From research and product decisions to working systems.
Projects
Multi-agent desktop workspace
CraftStation
Multiple AI coding agents, side by side, in one shared project context.
A cross-platform desktop workspace for running agent sessions in parallel, with built-in orchestration controls.
View case study
Research
Scaffolding Candidates from AI-Assisted Feedback and Dialogue to Independent Interview Practice
How AI-assisted interview coaching can support candidates without replacing the evaluative and productive work they need to perform themselves.
Human–AI interaction · AI feedback · Interview practice
Human–AI Collaborative Judgment of eWOM Authenticity for Purchase Decisions
How consumers and AI can jointly judge whether human-authored eWOM reflects genuine experience or promotional intent, while preserving uncertainty and user control.
eWOM · Human–AI judgment · Uncertainty
Dynamic Representation Augmentation for Multimodal Emotion Recognition with MLLMs
How multimodal emotion recognition can keep short-term visual and acoustic dynamics explicit, instead of leaving a language model to reconstruct them.
Multimodal emotion recognition · Dynamic representation · Audio–visual learning
About Me
I work across product discovery, AI system design and implementation.
Product discovery & problem framing
Turn user behavior, research evidence and product constraints into clear product decisions.
AI product definition & evaluation
Define AI workflows, feedback loops, guardrails and evaluation criteria for human–AI systems.
Cross-functional prototyping & shipping
Work closely with design and engineering, prototype when needed, and stay involved through implementation.
Blog
个人主页设计理念
shucubed.xyz 是一个关于维度、变化与映射的空间，也是我对产品、研究与工程三个坐标轴的一次阶段性映射。
用好 AI 的三步方法：拆解、杠杆、打磨
AI 工具不断变化，但决定结果的往往不是工具本身，而是怎么想、怎么选、怎么做：拆解问题、选择杠杆，然后持续打磨。
AI Coding 工作流 v3
一套按 Audit、Plan、Execute、Review、E2E、Commit 六个阶段推进的 AI Coding 任务生命周期，以及触发回归检查的条件与范围。
Claude Code 节约 Tokens 技巧
24 个 Claude Code 成本控制技巧：保护 prompt cache、subagent 隔离上下文、确定性任务脚本化、按难度切换模型、/compact 与 /clear 的时机、/rewind、/btw、/context 监控、Batch API、权限分层、git worktree 并行、adversarial review、失败模式库与 CodeGraph MCP。