Reform

We analyze the world’s best web apps, extract their UX patterns, rebuild your interface automatically, and simulate user behavior to prove the improvement.

Comment 1

🚀 About the Project

We built Reform to solve a problem we kept running into:
improving UI/UX is slow, subjective, and hard to validate before shipping.

Designers rely on inspiration, developers rely on guesswork, and teams often don’t know if a UI change will actually improve user behavior until after deployment.

We wanted to change that.

💡 Inspiration

We were inspired by how tools like GitHub Copilot automate coding — but there’s no equivalent for UI/UX.

At the same time, the best products in the world (like GitHub, Vercel, Linear) already solved many UX problems.
So we asked:

What if we could learn from the best interfaces, apply those patterns automatically, and predict how users will react before shipping?

That idea became Reform.

🏗️ How We Built It

Reform is a multi-stage AI system:

  1. Project Discovery

    • User describes their product
    • AI finds ~50 top competitors and selects the most relevant ones

  2. Real UX Analysis (TinyFish)

    • Browser agents visit each site
    • Simulate real user behavior (scrolling, clicking, navigating)
    • Extract structured UX and design data

  3. Design Intelligence Synthesis

    • AI merges insights into a unified design system:
      • layout patterns
      • components
      • design tokens (colors, spacing, typography)
      • UX flows

  4. UI Transformation Engine

    • Applies improvements in a strict order:
      • layout → spacing → components → visual polish
    • Maps everything to Tailwind-compatible values

  5. Code Generation

    • Outputs production-ready React + Tailwind code

  6. Simulated User Heatmaps (Key Feature)

    • 50 synthetic agents test both the original and improved UI
    • Generates:
      • attention heatmaps
      • engagement predictions
      • drop-off risk
    • Creates a before vs after behavioral comparison

🧠 What We Learned

  • Good UX is not just visual — it’s behavioral
  • The best products share consistent, extractable patterns
  • Simulating users is far more powerful than just analyzing layouts
  • AI becomes much more valuable when it closes the loop:

design → code → user behavior → feedback

⚔️ Challenges We Faced

  • TinyFish integration

    • The SDK had issues with streaming responses, so we had to bypass it and implement direct SSE parsing

  • Getting structured output

    • Ensuring consistent JSON across different websites required strict prompting and normalization

  • Balancing automation vs control

    • We needed deterministic transformation rules while still leveraging AI flexibility

  • Making results usable

    • It’s easy to generate ideas — much harder to generate code-ready, implementable outputs

🔥 What Makes Reform Different

Reform is not:

  • a UI generator
  • a theme switcher
  • a design inspiration tool

Reform is:

  • a UX intelligence engine
  • a UI refactoring system
  • a simulated user testing lab

Most tools say:

“Here’s a better design”

Reform shows:

“Here’s a better design — and here’s how user behavior will change.”

🎯 Final Thought

Our goal is to make UI improvement:

  • automatic
  • measurable
  • fast
  • data-driven

Instead of guessing what works, developers can see the impact before they ship.

Built With

Share this project:

Updates