AI UX Audit Tool
aiuxdesign.guide — Upload a screenshot of your AI product. Get a graded report on what's working, what's missing, and what to ship next — scored against 36 research-backed AI UX patterns.
What the audit does
Drop a screenshot (chat thread, agent surface, AI-powered feature) into the audit tool. The pipeline returns:
- A score out of the patterns that apply to your surface (typically 3–8 out of 36 — most surfaces don't meaningfully invoke all 36)
- Top gaps, ranked by severity, each with cited visible evidence from the screenshot
- A consolidated handoff prompt you can paste into Claude Code, Cursor, or another coding assistant — it locates the affected files in your repo and applies the patterns
- Per-gap pattern links for designers who want to go deep on one issue
The audit is free, no signup required for the first run.
How the grading works
The audit is run by Claude (Sonnet, temperature: 0) against a structured prompt that encodes all 36 patterns. The process is intentionally evidence-first, not pattern-first — earlier versions that led with "check pattern X" produced fabricated findings on surfaces where pattern X didn't apply. Current methodology, in order:
- Describe each surface. Identify what specifically is shown (e.g., "Claude chat thread mid-conversation", "Settings → Usage page"). Inventory the visible interactive controls. This list anchors every later "X is missing" claim — if a control wasn't in the inventory, claiming it's absent is hallucination.
- Select applicable patterns (max 8). From the 36-pattern library, list only the patterns that meaningfully apply to this specific surface. A billing page does not need Confidence Visualization; a chat thread does. Most surfaces invoke 3–7 patterns, not 20. Padding destroys trust in the audit, so the prompt has a hard cap.
- Evaluate only the applicable patterns. For each one, decide whether the surface implements it. Every finding must include an
evidence field that cites a specific visible UI element ("no thumbs-up/down on AI messages in the message list"). Findings without specific evidence get dropped — they are not invented.
- Rank gaps + propose actions. Each gap gets a severity (critical / high / medium / low) and a "what to ship" suggestion grounded in the pattern's install guidance.
The honest-fallback case matters as much as the happy path: if the screenshot isn't an AI product surface (e.g., a marketing page, a non-AI billing screen), the audit returns 0 applicable patterns and a useful empty-state explanation — not 8 fabricated gaps.
The 36 patterns powering the audit
The pattern library is the methodology — open source, with interactive demos and install prompts for each. Patterns are organized across 8 categories:
| Category | Patterns | Examples |
|---|
| Human-AI Collaboration | 10 | Autonomy Spectrum, Intent Preview, Human-in-the-Loop |
| Trustworthy & Reliable AI | 8 | Trust Calibration, Explainable AI, Safe Exploration |
| Adaptive & Intelligent Systems | 4 | Adaptive Interfaces, Ambient Intelligence |
| Natural Interaction | 4 | Conversational UI, Progressive Disclosure |
| Safety & Harm Prevention | 4 | Crisis Detection, Vulnerable User Protection |
| Performance & Efficiency | 3 | Intelligent Caching, Agent Status Monitoring |
| Privacy & Control | 2 | Privacy-First Design, Selective Memory |
| Accessibility & Inclusion | 1 | Universal Access Patterns |
Each pattern page includes: an opinionated judgment call ("use it when / don't when / the trap"), 4 ranked takeaways, real-world examples from shipped products, an interactive demo, and an "install with Claude Code" prompt that drops the pattern into your codebase.
Also at aiuxdesign.guide
- Designer Guides — structured learning paths covering AI UX topics
- Daily Newsletter — curated AI UX news; pattern-tagged so updates surface relevant changes to the library
- Prompt Library — reusable prompts for AI-powered design workflows
Tech Stack
| Layer | Technology |
|---|
| Framework | Next.js 15 (App Router) |
| UI | React 19, TypeScript |
| Styling | Tailwind CSS v4 |
| Database | PostgreSQL on Neon (Prisma ORM) |
| AI | Claude API (Anthropic SDK) — Sonnet for audit, Haiku for newsletter |
| Email | Resend (transactional) + Beehiiv (newsletter delivery) |
| Analytics | Microsoft Clarity, Vercel Speed Insights |
| Deployment | Vercel |
Getting Started
git clone https://github.com/imsaif/aiex.git
cd aiex
npm install
cp .env.example .env.local # configure environment variables
npx prisma generate
npm run dev
Open http://localhost:3000.
Environment Variables
See .env.example for the full list. At minimum:
DATABASE_URL — PostgreSQL connection stringANTHROPIC_API_KEY — for audit, mentor, and newsletter generationRESEND_API_KEY — for transactional emailBEEHIIV_API_KEY + BEEHIIV_PUBLICATION_ID — for newsletter subscriber sync
Development
npm run dev # Dev server (Turbopack)
npm run build # Production build
npm run lint # ESLint
npm test # Run tests
npm run test:coverage # Tests with coverage report
Contributing
Contributions welcome — new patterns, improvements to existing ones, audit prompt refinements, bug fixes, docs. See CONTRIBUTING.md.
License
MIT — see LICENSE.
Contact
