Normalize — Asif Mukthar
Side Quest · Figma Plugin · In Development

Normalize

A Figma plugin that audits and restructures existing design system components into an AI-LLM readable format. Fewer hallucinations, better component consistency when feeding your system to any AI.

Status
In Development
Type
Figma Plugin
Platform
Figma
Contact

Design systems weren't built for AI

Most design systems were built before AI-native workflows existed. Layer names like "Group 47" or "Rectangle 12" made sense to humans who could see the canvas. To an AI or LLM trying to parse, interpret, or generate from your components, they're meaningless noise.

The result: when you feed your design system to an AI tool, it either produces generic output that ignores your components, or actively hallucinates because it can't accurately read the structure. The problem isn't the AI. It's how your components are built.

AI-readiness at the component level

Normalize is a Figma plugin that audits your existing components and restructures them to be AI-LLM readable. It fixes layer naming conventions, nesting hierarchy, icon labeling, and component architecture, so that when AI reads your design system, it reads it accurately.

Audit

Component Analysis

Scans layer naming, nesting depth, icon labeling, and variant structure to identify exactly what is blocking AI comprehension.

Restructure

Auto-Fix Structure

Proposes semantic layer naming, proper nesting hierarchy, and component metadata aligned to AI LLM standards.

AI-Ready

Consistency Check

Validates that components pass AI inspection. Fewer hallucinations, better consistency when feeding your system to any LLM.

Scope of the audit

Layer Naming Conventions

Checks whether layer names are semantic (e.g., "button/primary/label" vs "Rectangle 47") and readable by LLM parsers.

Nesting Hierarchy

Validates that nested groups, frames, and components follow a logical, AI-parseable structure without redundant wrapping.

Icon Labeling

Ensures icons are named and tagged in a way that AI can identify their purpose and role within the component.

Variant Structure

Checks that component variants are named consistently and follow a structure that AI can map to design tokens and states.

Component Metadata

Validates that components include the metadata (descriptions, links, properties) that LLMs rely on for accurate interpretation.

Pass/Fail Scoring

Each component gets a readiness score. You see exactly what passes, what fails, and what to fix, with priority ranking by impact.

Why I'm building this

"Design systems built before AI weren't structured for machine understanding. Normalize bridges that gap, translating existing components into AI-readable format without rebuilding from scratch."

As someone who works with AI daily in my design workflow, I kept running into the same wall: the AI couldn't reliably work with my existing components. The fix wasn't a new design system. It was a translation layer. That's Normalize.

Interested in this?

Normalize is actively in development. If you're a designer or team lead dealing with AI hallucinations on your design system, I'd love to talk through the problem with you.

Let's Talk ↗