Lyra Wang

Work

What has actually been built

This domain records where the real project activity happened: the problem being addressed, the approach taken, what changed over time, and the current state.

Project records

Each record states the problem, the practice, the current stage, and what evidence exists.

01

Meeting Intelligence Agent

会议分析智能体

A meeting analysis agent integrated with real collaboration systems.

Problem

Meetings generate decisions, action items, and context that are hard to capture, structure, and follow up on — especially across tools people already use every day.

Practice

Built and iterated a meeting analysis agent through at least two tracked phases. Integrated with Feishu (Lark) systems, practiced third-party Open API integration, studied official API documentation, and worked through authentication, interfaces, events, and node chains. Gradually formed an "official docs + AI" development approach.

Current Stage

Used in a real collaboration context and iterated through multiple phases.

Evidence

Integrated with Feishu (Lark) systems, across multiple tracked phases of iteration.

02

Retail Operations System

进销存管理系统

A multi-year real-world retail operations system — a practice ground for product engineering.

Problem

A physical retail business needs sales, purchasing, and inventory to work together reliably over years — not as a one-off CRUD exercise, but as a system that can grow, be maintained, and be trusted.

Practice

Continuously practiced data structure design, began thinking about State Ownership and module boundaries, introduced Multi-Agent collaborative development with Architect / Developer / Reviewer / QA roles, and started paying attention to reliability, maintainability, and iterability.

Current Stage

Multi-year project, continuously evolving.

Evidence

A real, multi-year project serving an actual retail business scenario. Multi-Agent collaboration (Architect / Developer / Reviewer / QA) is being practiced.

03

WeChat Publishing System

微信公众号内容与发布系统

An AI-native publishing workflow that reduces the friction of public output.

Problem

Publishing written work publicly involves a chain of discussion, drafting, formatting, version control, platform submission, review, and release — each step adds friction that stops ideas from being shared.

Practice

Built a pipeline: ChatGPT for content discussion and crystallization → formatted files → GitHub for version control → automated entry into WeChat Official Account backend → human review → publication. Keeps human review in the loop by design.

Current Stage

AI-native publishing workflow in use.

Evidence

Real pipeline from ChatGPT through GitHub to WeChat Official Account backend. Human review is intentionally retained.

How this area is organised

Work is grouped into two areas, so that projects are not treated as interchangeable merely because they are repositories.

Flagship Products

Representative real-world products and systems.

Nothing assigned to this area yet.

Projects are listed above while their grouping is being decided.

AI Infrastructure

Reusable systems or infrastructure supporting AI-native work.

Nothing assigned to this area yet.

Projects are listed above while their grouping is being decided.