AKASH VISHWAKARMAAI Full-Stack Developer
Back to work
/ 022025

DesignRip

AI-Powered Design Intelligence

01

The Problem

Design systems are usually documented by hand, if at all — engineers copy hex codes and spacing values out of DevTools one element at a time. DesignRip asks a different question: given any public URL, can a machine extract the whole design system in the time it takes to load the page?

02

Approach

  1. 01

    A Playwright headless-browser pipeline visits the target URL, walks the rendered DOM, and captures computed CSS (color, type scale, spacing, radii, shadows) alongside high-fidelity component screenshots.

  2. 02

    Raw scraped JSON is passed to Groq running Llama 3.3-70B, which restructures it into a clean, human-readable design.md — palettes, type ramps, spacing tokens and component notes, ready to drop into a repo.

  3. 03

    Built on Next.js 15 for a fast, server-rendered intake flow (paste a URL, get a document) with minimal client-side weight.

  4. 04

    Optimized the extraction pipeline for speed — the Groq/Llama inference step is the only variable-latency stage, everything else runs in parallel where possible.

03

Gallery

01
Coming soonScreenshot — Overview
02
Coming soonScreenshot — Dashboard
03
Coming soonScreenshot — Detail view
04

Stack

Next.js 15PlaywrightGroqLlama 3.3-70B
05

Results

0countries
$0marketing
secto docs
Back to work