Building AI products, from a concrete problem to a working prototype.
I’m interested in the early decisions that shape a product: whose problem to solve, what the smallest useful experience looks like, and what evidence would justify the next step. I use code to explore those decisions through agent systems, human–AI interaction, and data tools.
I’m looking for Founder’s Office / founding-team and Product / Product Engineering internships at AI and intelligent-hardware startups. I’m also interested in product development and the path from prototype to adoption.
Each repository explains the intended experience, implementation, verification path, and current limits. These are prototypes and reference implementations; test results are separate from customer adoption.
| Project | Product question | What to inspect |
|---|---|---|
| CLAW — education agent | How can parents and teachers use a shared assistant while keeping each child’s records appropriately scoped? | Role-scoped reads and writes, persistent learning observations, and an OpenClaw integration harness. |
| DataSteward | What controls are needed before an AI agent can help with operational data? | A synthetic manufacturing-data workflow, governed tools, access policies, action previews, and rollback. |
| Cyber-Superego | How could a desktop assistant notice a distraction without making its own interventions disruptive? | Local visual perception, contextual classification, agent memory, and opt-in desktop actions. |
| Wren | How can an AI character maintain continuity while preserving boundaries and a distinct voice? | A Telegram relationship-simulation prototype with separate decision and speech stages, memory, and an offline evaluation harness. |
| Minecraft AI Companion | How can a game companion combine task execution with conversation? | Separate task and conversation agents connected to a Mineflayer control bridge. |
- Problem framing: connect a user scenario to a concrete product scope and a testable prototype.
- Hands-on implementation: work across Python, TypeScript, agent orchestration, APIs, and data workflows.
- Product tradeoffs: make privacy, permissions, failure behavior, and human control part of the experience.
- Evidence and iteration: distinguish what is implemented, what has been tested, and what still needs real-world validation.
I’m continuing to learn how to turn technical prototypes into products people choose to use. The repositories above are the clearest record of that work so far.
我关注 AI 产品从想法到落地的过程：先理解具体用户和场景，再用原型检验产品选择，并认真处理权限、隐私与失败时的体验。
目前优先寻找优秀 AI / 智能硬件创业公司的创始团队、Founder’s Office、产品与产品工程实习。我希望既能参与问题定义和产品判断，也能亲手把方案做出来，并继续学习用户验证、商业化和团队协作。
上面的项目分别探索教育、企业数据、桌面助手和交互式 Agent。各仓库保留具体实现、验证方法和当前限制；我也还在边做边学。