CloseLoop

• 

  CloseLoop’s English home screen — turning messy meeting notes into human-confirmed work.

• 

  The human review step: CloseLoop extracts decisions and proposed actions, while people keep final control.

• 

  Confirmed commitments become a lightweight tracker with owners, deadlines, and a ready-to-send follow-up.

Inspiration

Most meeting tools stop at a summary. But a summary does not create accountability: decisions get buried, owners are unclear, and follow-ups never happen.

We built CloseLoop to answer a simple question: how can an AI agent help a team move from a messy conversation to work that people actually agree to do?

What it does

CloseLoop turns meeting notes into a human-confirmed follow-up workflow.

A user can load or paste meeting notes. CloseLoop identifies the key decision and proposes action items with owners and deadlines. Nothing is silently assigned: a person reviews, edits, and explicitly confirms each commitment. Confirmed work then appears in a lightweight tracker, and the team can copy a ready-to-send meeting follow-up.

How we built it

CloseLoop is a React and TypeScript web application built with Vite. It uses deterministic meeting scenarios for a reliable live demo, with a clear agent workflow:

1. Read meeting notes
   
2. Extract decisions and proposed action items
   
3. Let a human review and confirm commitments
   
4. Track completion and generate a follow-up brief
   

We used Codex and GPT-5.6 throughout the project: to refine the product scope, build the React components and interaction flow, add tests for task state transitions, improve the multilingual interface, and prepare deployment.

The app is deployed at loop.shaokai.men, and the source is available on GitHub.

Challenges we ran into

The hardest product question was not extraction—it was trust. An agent can propose tasks, but it should not quietly create commitments on behalf of a team. We designed the confirmation step as a first-class part of the product: people remain responsible for the final decision.

We also needed the demo to feel reliable and easy to understand in a short time. That led us to use structured scenarios and focus the MVP on one complete loop instead of trying to solve every meeting workflow at once.

Another challenge was making the experience work naturally in both Chinese and English. We added automatic system-language detection, a manual language switcher, localized meeting scenarios, and consistent typography across both interfaces.

What we learned

We learned that the most useful AI agent is not one that does everything automatically. It is one that reduces the tedious work, clearly shows its reasoning, and leaves meaningful decisions to people.

CloseLoop is our first step toward meetings that produce momentum—not just another document nobody revisits.

Built With

• agents
• ai
• codex
• gpt-5.6
• human-in-the-loop
• management
• meeting
• multilingual
• notes
• productivity
• react
• task
• typescript
• vite