Others learn to talk to AI. I learn to run it.
I build practical AI products that move beyond demos: agentic workflows, evaluation pipelines, domain-specific skills, and full-stack prototypes that can survive real product constraints.
- 🧭 Product: turning ambiguous problems into testable AI product loops
- 🤖 Agents: multi-agent orchestration, skills, tool use, quality gates, and human review
- 🧪 Evaluation: benchmark design, failure analysis, acceptance criteria, and iteration
- 🛠️ Building: Python, TypeScript, JavaScript, Next.js, NestJS, and lightweight local tools
- 🌏 Interests: AI-native products, knowledge tools, content systems, ASR, RAG, and GEO/SEO
| Project | What it does |
|---|---|
| SEO Crew Agent | Multi-agent B2B SEO pipeline: vision analysis → research → planning → writing → quality review, with revision loops and WordPress-ready output. |
| Elec Sheep Longform | Codex/Claude skill for producing consistent 8-page Xiaohongshu educational carousels with an IP character, editable SVG text, batch rendering, and visual QA. |
| ASR Transcription Skills | A field-tested skill suite for transcription pipelines, engine benchmarking, SRT post-processing, and learning-note generation. |
| doc2learn | Zero-dependency generator that turns Markdown/TXT course material into interactive, offline learning pages with experiments and checks. |
| DocGate | Local Markdown change-acceptance middleware that separates human feedback, agent claims, real diffs, and final review decisions. |
| Buffet Lens | Full-stack buffet decision and contribution MVP built around traceable restaurant, menu, evidence, and review data. |
Building AI products that are useful, testable, and ready to be operated.