Unity Atlas - AI Design System Assistant
Turning scattered design system documentation into instant, cited answers through semantic search and AI.
The Problem
The Problem
Our design system docs were scattered across Figma, Confluence, etc. Simple questions like 'What's our primary green?' or 'What's the spacing scale?' took 5-10 minutes of hunting. Watching peers waste time digging through files every day got old fast.
The Solution
The Solution
I built Unity Atlas to fix that. Ask questions in plain English; get instant answers with citations to the source documents. Semantic search + LLMs + a few weekends of coding. Not revolutionary. Just useful.
See It In Action
See It In Action
See it in action. Responses blurred for confidentiality, but the interface speaks for itself.
Technical Approach
Technical Approach
Built on RAG architecture: semantic search + vector embeddings + LLMs. The system ingests our design system docs, chunks them, creates embeddings, and then retrieves relevant context when you ask questions. Every answer includes citations so you can verify against the source. Tech: Next.js, TypeScript, Vercel AI SDK, vector search. Secure auth, real-time streaming.
Results & Impact
Results & Impact
Unity Atlas is already showing significant time savings, especially when it comes to newly onboarded team members. v1 focused on onboarding and filling knowledge gaps; this sits alongside our docs, doesn't replace them.
