RepairSync

A neutral AI referee for auto-repair disputes: audits collision estimates against live parts pricing, flags overcharges and skipped safety steps, and issues a report all sides trust.

Comment

3rd place Claremont Accelerator Hackathon

RepairSync

The Neutral AI Referee for Auto Repair & Insurance

License: MIT Stack: Turborepo Backend: Convex

RepairSync is a "Truth Layer" for the auto insurance ecosystem. It uses computer vision, real-time pricing intelligence, and clinical-grade reporting to resolve friction between insurers, repair shops, and drivers.

🛡️ Core Features

1. Dispute Shield™

The Dispute Shield acts as an automated auditor for collision estimates. It compares manual shop estimates against AI-driven "Referee Reality" to identify:

  • 🔴 Overcharges: Pricing that exceeds fair market value (>10% delta).
  • 🟡 Safety Risks: Critical procedures (like ADAS calibrations or sensor checks) that were omitted by the shop.
  • 🟢 Agreed Items: Line items that match market standards within a 5% tolerance.

2. Parts Pricing Pipeline

A sophisticated 3-stage engine that ensures precision in part valuation:

  • Stage 1: VIN-to-SKU: Decodes VINs via NHTSA and maps damage to exact OEM part numbers.
  • Stage 2: Multi-Source Scraper: Real-time pricing from OEM Dealers, Aftermarket (RockAuto), and Salvage (Car-Part.com) using Playwright stealth automation.
  • Stage 3: Normalization: Handles shipping, core charges, and interchange compatibility to provide a "Total Landed Cost."

3. Referee Reporting

Generate clinical, "Neutrality Certified" PDF reports designed to be the final word in indemnity disputes. Reports include:

  • Vehicle IQ (VIN decoding & severity matrix).
  • Visual Evidence (AI-rendered bounding boxes on impact zones).
  • Dispute Ledger (Transparent Side-by-Side comparison).

🚀 Tech Stack

  • Monorepo: Turborepo
  • Backend: Convex (Real-time DB, File Storage, Edge Functions)
  • Web: Next.js 15, shadcn/ui, Tailwind CSS v4
  • Mobile: Expo / React Native
  • AI: Claude 3.5 Sonnet (Vision + Reasoning), OpenAI Whisper
  • State: Zustand
  • PDF Engine: @react-pdf/renderer

📂 Repository Structure

repair-sync/
├── apps/
│   ├── web/              # Next.js — Shop/Insurer dashboards
│   ├── mobile/           # Expo — Driver photo capture & instant estimates
│   └── convex/           # Backend functions, schema, and AI logic
├── packages/
│   ├── core/             # Pricing formulas, scrapers, and CV constants
│   ├── ui/               # Shared Tailwind/NativeWind components
│   ├── types/            # Zod schemas & TypeScript interfaces
│   └── config/           # Shared ESLint, TSConfig, and Tailwind configs
└── turbo.json            # Monorepo orchestration

🛠️ Getting Started

Prerequisites

Installation

npm install

Environment Variables

Create a .env.local in the root (or set in Convex dashboard):

ANTHROPIC_API_KEY=your_key
OPENAI_API_KEY=your_key
PARTSVOICE_API_KEY=your_key
PROXY_SERVER=your_proxy

Development

Start the full stack (Web, Mobile, and Convex):

npm run dev

⚖️ Our Philosophy: The "Referee" Persona

Unlike traditional estimating platforms that act as a workspace for negotiation, RepairSync provides a neutral audit. Our tone is clinical, objective, and data-driven—we don't take sides; we defend the data.

For detailed implementation specs, see the /docs directory.

Built With

Share this project:

Updates