I build practical AI products and automation systems with runnable demos, clear docs, safety gates, and CI-backed delivery.
status public repositories stars followers profile views X V2EX
📦 Repositories · ⭐ Portfolio Highlights · 🧰 What I Build · 🧭 How I Work · 📬 Contact
These are the repos I would point a recruiter, collaborator, or technical reviewer to first.
| Project | What it proves | Evidence |
|---|---|---|
| PromptPanel | macOS product thinking, local-first UX, release-readiness discipline | Swift app, screenshots, CI release-readiness check, user-facing README |
| telegram-ui-builder | frontend product execution for bot/workflow builders | Live GitHub Pages demo, React/Vite app, Pages deploy CI |
| bilibili-cleaner | risky-account automation with review gates and API/CLI surfaces | FastAPI, Web UI, Typer CLI, pytest CI, rate-limit and safety notes |
| x-account-cleaner | browser automation for destructive account cleanup with safer workflows | TypeScript, Playwright, CI, review-first cleanup model |
| ecommerce-product-image-workflow | AI workflow packaging for non-engineering users | Python MVP CLI, templates, QA docs, GitHub Pages |
| reality-resi-stack | operator-facing infra docs and deployment tooling | Shell toolkit, bilingual runbook, redaction CI, GPL-3.0 |
name: qilai role: AI Product Engineer · Full-stack Builder · Agent Harness Builder location: China · GMT+8 focus: - AI products that turn prompts and agents into repeatable workflows - Agent harnesses with tools, memory, evals, logs, and operator controls - Local-first automation with dry runs, review gates, limits, and recovery notes - Developer tools, CLIs, bot builders, and operations-facing utilities strengths: - idea -> runnable MVP -> docs -> CI -> iteration - product judgment + harness engineering + runtime safety - frontend + backend + automation + packaging - making risky workflows reviewable instead of opaque principles: - "Make the useful path work first." - "Keep failure modes visible." - "Ship small, learn fast, improve from evidence." open_to: - AI product engineering roles - Practical AI products - Agent harness R&D / productization - Developer automation - Local-first open-source tools
- 🛠️ Polishing PromptPanel — a macOS prompt and snippet launcher for AI power users
- 🧪 Exploring agent harness patterns — tool use, memory, evals, logs, guardrails, and human-in-the-loop controls
- 🧹 Maintaining account/workflow automation tools such as bilibili-cleaner and x-account-cleaner
- 🎨 Improving developer-facing builders such as telegram-ui-builder and Codex workflow utilities
- 📦 Raising the quality bar of public repos: README, quick start, CI, safety boundaries, licenses, topics, demos, and Star History
| Direction | What I care about | Public evidence |
|---|---|---|
| AI products | Turning prompts, agents, and LLM workflows into usable tools | PromptPanel, ecommerce-product-image-workflow |
| Agent harness | Productizing agents with tool contracts, eval loops, memory, observability, and operator control | PromptPanel, Codex workflow utilities, automation repos |
| Automation | Local-first workflows, dry runs, review steps, logs, limits, and rollback notes | bilibili-cleaner, x-account-cleaner |
| Developer tools | CLIs, bot builders, operator consoles, and workflow kits | telegram-ui-builder, codex-app-account-switcher |
| Infrastructure utilities | Small tools with clear runbooks and operational guardrails | reality-resi-stack, IDM-Activation-Script-Chinese |
- 🎯 Find the real workflow — user entry point, critical path, failure mode, and reason to return.
- 🧪 Design the harness, not just the prompt — tools, state, evals, logs, permissions, and recovery paths.
- 🧱 Build the smallest useful loop — runnable first, then polish from actual friction.
- 🛡️ Harden risky actions — dry runs, confirmations, logs, limits, and recovery notes.
- 📚 Package for the next visitor — README, quick start, status, license, screenshots or demo where useful.
- 🔁 Iterate from evidence — issues, runtime failures, user confusion, stars, and repeated usage.
Primary stack
TypeScript / Python / React / FastAPI / PostgreSQL / Docker / OpenAI · Claude / Agent Harness / Local-first Automation
Languages
Frontend
Backend & Data
AI / LLM
OpenAI Claude LangChain LlamaIndex Hugging Face Agents RAG Vector DB Prompt Systems Evals Tool Calling Memory Guardrails Observability
Agent Harness / AI Engineering
Agent Harness Agent Harness Product Agent Runtime Tool Contracts Function Calling MCP Workflows Human in the Loop Eval Harness Tracing Operator Console Product Loop Safety Boundary
DevOps & Tools
public repositories owner stars followers
GitHub Email X V2EX Open to practical collaboration
✨ Code is useful when it removes friction. Automation is valuable when it stays under control. ✨
Repository count is dynamic because the open-source portfolio keeps growing.