FairQuote

AI-powered auditor that detects overcharges and fraud in auto repair estimates, then generates data-backed dispute scripts — returning pricing power to consumers.

Comment

Inspiration

$70 billion is lost annually to auto repair overcharges. Mechanics exploit information asymmetry — the average consumer pays 30% more simply because they can't evaluate technical jargon. We asked: who audits the mechanic?

What it does

FairQuote lets consumers snap a photo of any repair estimate. The system:

  1. Parses the document using OCR
  2. Cross-references each line item against regional market labor rates
  3. Flags anomalies — overpriced parts, inflated labor, and outright fraud (e.g., billing for parts that don't exist)
  4. Generates a dispute script — a data-backed message ready to send to the repair shop

How we built it

We applied strict Systems Thinking to decouple the architecture into three layers:

  • Frontend Layer — Next.js 16 (App Router) with a 4-state React lifecycle (Upload → Processing → Results → Dispute), animated with Framer Motion spring physics
  • API Route — /api/analyze handles file validation and routing
  • AI Service Layer — A fully isolated mock service (lib/services/aiService.ts) designed for seamless replacement with real OCR (Google Vision) + LLM (GPT-4o) endpoints

For the UI, we engineered a premium Liquid Glass interface using SVG fractal noise (feTurbulence) and displacement mapping (feDisplacementMap) — a 3-layer compositing technique that creates true optical refraction without distorting text content.

Challenges we ran into

Our first approach — wrapping OCR output into a raw LLM prompt — failed. The AI hallucinated part numbers, and the chatbot-style UI destroyed consumer trust. This forced us to rethink the entire architecture: separating the AI service from the presentation layer, and investing in a premium UI that communicates institutional credibility rather than casual conversation.

The Liquid Glass effect also required solving a critical CSS layering problem: SVG displacement filters distort all children of a container, including text. We solved this with a 3-layer compositing architecture (distorted background → specular highlights → crisp content).

Accomplishments that we're proud of

  • Zero-dependency AI service swap: The mock service can be replaced with real APIs without touching a single UI component
  • Fraud detection: The system identifies non-existent parts (e.g., "Blinker Fluid") as fraudulent charges
  • Production-grade Liquid Glass: True SVG displacement + backdrop-blur compositing, not just CSS blur
  • One-tap dispute generation: Consumers get actionable leverage, not just information

What we learned

Trust in fintech isn't just about accuracy — it's about presentation. A chatbot UI, even with perfect analysis, fails to inspire confidence when people are dealing with potential scams. Enterprise-grade visual design is a functional requirement, not a cosmetic one.

What's next for FairQuote

  • Real AI integration: Swap the mock service for Google Vision OCR + GPT-4o analysis
  • Database persistence: Supabase/Postgres for estimate history and user accounts
  • WhatsApp API: Direct dispute script delivery to repair shops
  • Multi-language support: Spanish, Portuguese for Latin American markets

Built With

Share this project:

Updates