## Inspiration

Young people early in their careers carry a mild ongoing uncertainty — about who they are, what they're good at, where they're going. They don't want therapy. They don't want habit trackers. They want something that notices them without analyzing them.

微光 (Wēi Guāng, "faint light") came from that feeling. A quiet record written by someone who noticed you.

## What it does

微光 is a low-pressure AI companion. It has a short, gentle conversation with you — 1–2 sentences at a time, warm and reactive, never clinical. When the moment feels complete, it quietly writes a diary entry in its own voice: third-person, observational, warm. The entry ends with "在微光看来," — 微光 speaking as itself, offering its own quiet perspective on your day.

No account. No login. No cloud. Everything lives on your device. The diary card appears directly in the chat thread and can be saved as a photo to share on Xiaohongshu.

Note: DeepSeek API and Vercel are not accessible from mainland China without a VPN. A production version would use a China-accessible backend — this build is a hackathon demo.

## Challenges we ran into

This is the first time I've fully built a piece of software using Claude Code. I don't have much coding background, so it was hard to gauge how difficult certain things would be. There were quite a few challenges during debugging — getting the AI to return consistent JSON, tuning 微光's voice so it felt warm and not clinical, and figuring out how to export the diary card as a clean image. But we got through all of it. Thank you, AI, for the guidance.

## Accomplishments that we're proud of

The AI diary section. After just a short conversation, 微光 writes a third-person diary entry in its own voice — ending with its own quiet observation on your day. It's a lightweight product, but that moment lands in a way that surprised me. That's the heart of it.

## What's next for 微光 (Wēi Guāng)

The biggest next step is making 微光 accessible on the Chinese internet — finding a China-compatible AI backend so the people it was built for can actually use it. Beyond that: helping users surface their own strengths over time, callbacks to past conversations, and a browsable diary view. The foundation is there. The next version is about depth.

Built With

  • deepseek-api
  • localstorage
  • react
  • vercel
  • vite
Share this project:

Updates

Submission history