fix.autodun.com is a free, AI-powered vehicle fix assistant hub built for UK drivers. It combines Claude's language and vision capabilities with a clean, purpose-built interface to help drivers diagnose breakdowns, decode warning lights, appeal parking fines, and check whether a used car is fairly priced — all without needing to book a mechanic or pay for advice.
Every result includes a thumbs up / down feedback mechanism. That feedback is stored in Supabase as structured ML training data, building a dataset to train future Autodun AI models on real-world UK vehicle problems.
This project is built and maintained by a solo UK-based founder as part of the broader Autodun platform.
No login required. Works on mobile and desktop.
Describe any vehicle problem and receive a structured AI diagnosis in seconds.
- Input: vehicle make, model, year, mileage, driveability status, and problem description
- Output: likely cause, severity rating (Critical / High / Medium / Low), whether it's safe to drive, immediate action steps, what to tell your mechanic, and a UK repair cost estimate
- Powered by Claude with a domain-specific mechanic system prompt
Three input methods for maximum flexibility:
| Method | Description |
|---|---|
| 📷 Photo Upload | Upload a dashboard photo — Claude Vision identifies the light and explains it |
| ⊞ Grid Picker | Tap any of 16 common warning lights (oil, ABS, airbag, DPF, TPMS, etc.) |
| 🔌 OBD Code | Enter a fault code (e.g. P0300) for a full technical breakdown |
Output includes: what the light means, severity, whether it's safe to drive, likely causes, repair steps, UK cost estimate, and a mechanic tip.
Generates formal, ready-to-send UK appeal letters based on the circumstances of the fine.
- Supports both council PCNs and private parking charges
- Structured form: vehicle reg, date, location, reason on notice, grounds for appeal
- Output references relevant legislation: Traffic Management Act 2004, Protection of Freedoms Act 2012, BPA / IPC Codes of Practice, POPLA / IAS appeal procedures
- One-click Copy Letter button on the result
Two modes for evaluating a used car's price:
📋 Text Details mode:
- Paste any UK listing text for instant analysis
- Paste an AutoTrader / eBay Motors URL → a structured helper form appears (AutoTrader is login-protected; the form captures make, model, year, mileage, price, seller type, and extras)
- Output: GREAT DEAL / FAIR PRICE / OVERPRICED / UNDERPRICED verdict, green flags, red flags, fair market value, negotiation tips, and a pre-purchase checklist
📷 Photo Valuation mode:
- Upload any car photo (exterior works best)
- Claude Vision identifies the make, model, approximate year, and trim level
- Output: vehicle identified, condition rating (Excellent / Good / Fair / Poor), estimated UK market value range, key price factors, and a buying / selling tip
Every result screen includes a feedback bar:
"Was this helpful? Your feedback trains the Autodun AI" 👍 👎
- Thumbs up / down vote captured on click
- Optional free-text comment (submitted on Enter or Send)
- Each feedback row inserted into Supabase with:
tool,vote,note,result_summary(first 200 chars),created_at - Builds a labelled training dataset of real UK vehicle queries and AI responses for future fine-tuning
| Layer | Technology |
|---|---|
| Framework | Next.js 14 — App Router, Server Components, API Routes |
| AI | Anthropic Claude API (claude-sonnet-4-20250514) with Vision |
| Database | Supabase (PostgreSQL) — ML feedback storage |
| Deployment | Vercel |
| Styling | Pure CSS — custom design system, no Tailwind, no CSS-in-JS |
| Font | Inter via next/font/google |
| Token | Value | Usage |
|---|---|---|
| Background | #070f1a |
Page background |
| Card | #111f33 |
Surface / card background |
| Green | #00d48a |
Primary CTA, active states, section headings |
| Blue | #2979ff |
Focus states, selected items |
| Orange | #ff9500 |
Medium severity, warnings |
| Red | #ff4444 |
Critical severity, error states |
| Yellow | #ffd60a |
Low/medium severity badges |
- Node.js 18+
- An Anthropic API key
- A Supabase project with the
fix_feedbacktable (schema below)
# Clone the repository
git clone https://github.com/kamrangul87/fix-autodun-com.git
cd fix-autodun-com/app
# Install dependencies
npm install
# Copy the example env file and fill in your keys
cp .env.local.example .env.local
# Start the development server
npm run devOpen http://localhost:3000 in your browser.
Create a .env.local file in the app/ directory:
ANTHROPIC_API_KEY=sk-ant-...
NEXT_PUBLIC_SUPABASE_URL=https://your-project.supabase.co
NEXT_PUBLIC_SUPABASE_ANON_KEY=eyJ...| Variable | Required | Description |
|---|---|---|
ANTHROPIC_API_KEY |
✅ | Anthropic API key — used server-side only in /api/claude |
NEXT_PUBLIC_SUPABASE_URL |
✅ | Your Supabase project URL |
NEXT_PUBLIC_SUPABASE_ANON_KEY |
✅ | Supabase anon/public key for client-side inserts |
Note:
ANTHROPIC_API_KEYis accessed only in theapp/api/claude/route.jsserver-side route and is never exposed to the browser. The Supabase keys use theNEXT_PUBLIC_prefix because feedback is inserted directly from the client.
Run this in the Supabase SQL editor to create the feedback table:
create table fix_feedback (
id uuid primary key default gen_random_uuid(),
tool text not null,
vote text not null check (vote in ('up', 'down')),
note text,
result_summary text,
created_at timestamptz not null default now()
);User submits query
│
▼
Claude generates result (via /api/claude proxy)
│
▼
Result displayed to user
│
▼
User clicks 👍 or 👎 (optionally adds comment)
│
▼
Supabase insert → fix_feedback table
├── tool ("breakdown" | "lights" | "appeal" | "price")
├── vote ("up" | "down")
├── note (free-text comment or null)
├── result_summary (first 200 chars of AI response)
└── created_at (timestamp)
│
▼
Labelled dataset → future fine-tuning of Autodun AI
This creates a continuously growing, human-labelled dataset of:
- Real UK vehicle breakdown queries + AI diagnoses
- Warning light lookups + explanations
- Parking fine details + appeal letters
- Car listing analyses + price verdicts
fix.autodun.com (Next.js 14 — App Router)
│
├── app/
│ ├── page.js ← Single-page hub (client component)
│ │ ├── BreakdownAssistant
│ │ ├── WarningLightDecoder (photo / grid / OBD)
│ │ ├── ParkingFineAppeal
│ │ └── FairPriceChecker (text / photo)
│ │
│ ├── api/claude/route.js ← Server-side Anthropic proxy
│ │ └── POST /api/claude (keeps API key server-side)
│ │
│ ├── lib/supabase.js ← Supabase client
│ ├── globals.css ← Full design system
│ └── layout.js ← Inter font + metadata
│
└── Deployed on Vercel
Key architectural decisions:
- All Anthropic API calls are proxied through
/api/claude— the API key never reaches the browser - Vision requests (base64 images) are sent through the same proxy route, keeping the interface uniform
- Feedback inserts go directly from the browser to Supabase using the anon key (safe — row-level security can be added to restrict to inserts only)
- No global state management library — React
useStateper component is sufficient for this use case - No CSS framework — the design system is ~600 lines of hand-written CSS variables, utility classes, and component styles
Contributions, issues, and feature requests are welcome.
- Fork the repository
- Create a feature branch:
git checkout -b feature/your-feature - Commit your changes:
git commit -m 'add your feature' - Push to the branch:
git push origin feature/your-feature - Open a Pull Request
This project is licensed under the MIT License.
Kamran — solo founder and developer, based in the UK.
Part of the Autodun platform — building AI tools for UK drivers and the automotive industry.