A project-led engineering portfolio

Why it exists

MeteorTest exists to connect scattered test repositories, task scheduling, local execution, reports, and AI-assisted project operations into one practical platform loop.

What I built

• Next.js web console for projects, suites, tasks, builds, executors, reports, settings, and AI assistant surfaces.
• Python Local Agent for task polling, suite execution, log capture, and report upload.
• AI assistant surfaces for project import, suite import, task creation, report inspection, and failure analysis.

Role and validation results

Flagship platform and control plane.

Harden the live Web preview with public-preview mode, access protection, safe demo data, and clearer task/report surfaces.

Why it exists

MeteorVoice turns English practice into a spoken loop where listening, endpointing, AI feedback, and speech playback can be measured and improved together.

What I built

• Mobile and Web session surfaces with shared API client behavior, authentication, preferences, history, and settings.
• ASR provider layer with native STT fallback, Xunfei WebSocket routing, PCM capture diagnostics, and provider-level observability.
• TTS provider routing, session lifecycle controls, echo guard, endpointing, grouped loading feedback, and deployment paths for Tencent Cloud and Vercel.

Role and validation results

Voice-first learning product and cross-platform AI interaction system.

Improve bilingual response routing, voice quality, ASR/TTS diagnostics, and production deployment hardening.

Why it exists

This repository gives MeteorTest a real automation project to integrate with instead of only demonstrating platform screens.

What I built

• pytest-based API, iOS UI, and performance suite structure.
• Appium/XCUITest Page Object workflow and Allure report output.
• meteortest.yml contract used as the first platform integration sample.

Role and validation results

Companion test-code project and integration sample.

Keep the deterministic local mock API as the stable target for direct smoke execution and MeteorTest Agent validation runs.

Why it exists

A personal homepage gives the project family one stable public entry point instead of scattering context across individual repositories.

What I built

• Bilingual project showcase with localized routes and shared content modules.
• Flagship project framing for MeteorTest and related companion projects.
• Structured project pages that keep public project context easy to scan.

Role and validation results

Public project hub and portfolio site.

Keep the project list current as new repos land, and continue tightening the relationship between the homepage and each project detail page.

Why it exists

AI coding agents can read code but can't see UI. Lensport bridges this gap by providing screen context through the MCP protocol, running entirely locally.

What I built

• MCP server over stdio with screen capture, screenshot list, and recent screenshot tools.
• macOS and Windows screen capture with platform-native implementations.
• Local-first storage — screenshots stay on your machine, no cloud uploads.

Role and validation results

Local screen context MCP server for AI agents.

Expand platform support, improve capture performance, and add region-select and annotation tools.

Why it exists

Terminal AI agents can't see your screen, and manually saving, renaming, and uploading screenshots breaks the flow. AgentShot turns clipboard screenshots into instant AI context.

What I built

• Cross-platform CLI for macOS, Windows, and WSL.
• Clipboard-to-prompt bridge with automatic file naming and storage.
• One-command install scripts for Bash and PowerShell.

Role and validation results

Screenshot-to-prompt bridge for terminal AI agents.

Expand agent format support and add more platform integrations.

Why it exists

Operational systems work better when attendance, device, and session data are structured instead of scattered in ad hoc records.

What I built

• REST APIs for registration, login, device reporting, and session queries.
• Docker Compose setup for backend, MySQL, and Redis.
• Simple service-layer structure for model, repository, controller, and security code.

Role and validation results

Utility backend service.

Expand the reporting and workflow layer if it becomes part of a larger product system.

Why it exists

Many AI builds fail because the idea is not shaped well enough before implementation starts.

What I built

• Explore, package, and skill-mining modes for shaping delivery workflows.
• Spec, plan, one-shot prompt, handoff, chronicle, and skill suggestion outputs.
• Codex and Claude Code installation guidance for reusable workflow packaging.

Role and validation results

Reusable AI workflow package.

Keep tightening the workflow packaging, documentation, and reusable skill suggestions.

Why it exists

Agent tools are powerful, but project work still needs constraints around scope, validation, and user-owned changes.

What I built

• Codex-compatible SKILL.md guardrails.
• Claude Code guidance snippet.
• Installation and repository packaging for reuse across projects.

Role and validation results

Agent workflow quality layer.

Use it as a reference for future project AGENTS.md files.

Why it exists

Creator workflows benefit from repeatable video processing steps before a human reviews and publishes the final draft.

What I built

• FFmpeg scripts for replay download, clipping, vertical conversion, slow motion, subtitles, and BGM mixing.
• Bash and PowerShell variants for cross-platform use.
• OpenClaw/Codex/Claude Code guidance for reviewed publishing and copyright-aware preparation.

Role and validation results

Separate creator tooling project.

Improve reviewed draft preparation guidance and keep source-material usage clear.