Back to work
AI ApplicationAgent: VL-Coresystem

Demo · AI Title Lab

Demo case: an AI title tool for Xiaohongshu creators — showing the Problem → Call → Build → Outcome format.

Problem

Xiaohongshu creators spend 20–40 minutes a day on titles: write five by gut feeling, post one, and gamble. Titles drive most of the open rate, yet the process is pure superstition.

My call

  • No "universal copywriting tool": solve exactly one action — paste note content, get usable titles back
  • Constrain the model with a library of proven title formulas instead of letting it freestyle — controllable beats clever
  • Web app first, no browser extension: creators already organize material in the browser

How it was built

Paste note contentMatch formula libraryGenerate 10 candidatesRank by estimated open rate

Three implementation calls that mattered:

  1. Formula-first: every title is labeled with the formula used and why it fits
  2. Refuse to bluff: if the note lacks substance, the tool says "add a selling point first" instead of forcing output
  3. One-click copy: output is the end of the flow — zero extra dwell time

Outcome

  • Title time dropped from 20–40 minutes to 2 minutes
  • 12 creators in the pilot, 10 still using it the next day
  • The four-section structure of this page was inserted with the editor's one-click "Case study" template