Building AI-native products, agent loops, and harness layers that turn models into reliable systems.
I'm the co-founder of Mew.Design. I build AI-native products where the hard part is not the prompt itself, but the loop around it: planning, tool use, memory, feedback, editing, evaluation, and production reliability.
My background is product management, but my current work is much closer to code. I use AI coding agents daily to design workflows, ship features, debug systems, and turn rough product ideas into working software.
| Founder / operator | Agent loop products | Harness / framework practice |
|---|---|---|
| I turn ambiguous opportunities into shipped products, user loops, and business outcomes. | I design workflows where models plan, call tools, produce editable outputs, and improve through feedback. | I build around context, memory, permissions, observability, evaluation, deployment, and real-world constraints. |
I contribute to and learn from open-source AI agent, AI coding, and agentic workflow projects:
| Project | Why it matters to my work |
|---|---|
| Mastra | TypeScript agent framework for building AI applications, workflows, memory, tools, and observability. |
| OpenHands | AI software engineering agent and runtime environment for delegated coding work. |
| Dify | Agentic workflow platform for building and operating AI applications. |
| Project | What it is | My focus |
|---|---|---|
| Mew.Design | AI-native creation platform for generating and editing visual work. | Founder, product direction, agent workflow design, AI integration, core product development. |
| screenshot | Browser rendering service for converting HTML into production-ready images. | Rendering infrastructure, browser automation, cropping, optimization, and service reliability. |
| MagoUI | Local prompt-to-UI prototype experiment. | GenUI exploration, interface generation, product prototyping, and agent loop design. |
| BaiQuant-public | A-share quantitative research toolkit built with AI-assisted workflows. | Harness-style backtesting, stock selection, paper trading, and risk review loops. |
- Building Mew.Design as a practical AI-native product, not a demo.
- Designing agent loops for creation, editing, review, and iteration.
- Building harness layers around models: tools, memory, permissions, context, evaluation, and feedback.
- Contributing to open-source agent frameworks and agentic workflow projects.
- Using AI coding agents to ship real product work with tighter build-review-release loops.
