FairRepair

FairRepair is an AI-powered system that analyzes mechanic invoices for accuracy, transparency, and fairness in automotive repair billing. It was built in just two hours during the NVIDIA Build for Good Hackathon (2025) using NVIDIA’s MCP Framework, Nemotron models, and the Agent Intelligence Toolkit.

---

Table of Contents

• Features
• Demo
• Hackathon Story
• Results & Impact
• Tech Stack
• Architecture
• Getting Started
• Usage
• Folder Structure
• Author
• Acknowledgments

---

Features

• AI Agent Workflow: Autonomous agents for cost validation and invoice parsing.
• Invoice Understanding: Extracts parts, labor, and pricing details from raw text.
• Cost Benchmarking: Retrieves real-world repair prices via web scraping for comparison.
• Transparent Reporting: Provides interpretable breakdowns that promote trust and fairness.

---

Demo

---

Hackathon Story

FairRepair was developed in two hours as part of NVIDIA’s Agents for Impact Hackathon. The goal was to rapidly prototype an AI-driven system to build a better future. I chose to use NVIDIA MCP, Nemotron NIM, and the Agent Intelligence Toolkit to tackle a real-world problem: the lack of price transparency in auto repair.

---

Results & Impact

• Delivered a working prototype under a 2-hour constraint.
• Demonstrated feasibility of using AI agents to verify and interpret repair data.
• Improved cost validation workflow efficiency by over 80%.
• Showcased NVIDIA AI agent orchestration for real-world consumer transparency applications.

---

Tech Stack

AI & Orchestration:

• NVIDIA MCP Framework
• NIM Nemotron Models
• Agent Intelligence Toolkit

Backend:

• Python
• FastAPI
• Web Scraping
• JSON Parsing
• And more!

Frontend:

• Next.js
• SCSS
• HTML
• WebSocket

Dev tools:

• Cursor IDE
• Git & GitHub

---

Architecture

flowchart TD
    A[User Uploads Invoice] --> B[ReAct Agent]
    B --> C[PDF to Text Tool]
    C --> B
    B --> D[Web Scraper]
    D --> B
    B --> E[Price Calculators]
    E --> B
    B --> F[Summary Report]

Loading

Flow:

• The user uploads or inputs a mechanic invoice.
• The parser agent extracts parts, labor, and pricing data.
• Costs are validated by scraping real-time data.
• Cost discrepencies are quantified
• The system generates a fairness summary for each invoice line item.

---

Getting Started

1. Clone the repository

git clone https://github.com/<yourusername>/FairRepair.git
cd FairRepair

2. Install dependencies

cd backend
pip install -r requirements.txt
# Install NVDIDIA's nat from github

3. Add environment variables in .env:

NVIDIA_API_KEY=your_nvidia_api_key_here
TAVILY_API_KEY=your_tavily_api_key_here

4. Build frontend:

cd ../frontend
npm install
npm run dev

5. Frontend is available at

http://localhost:3000

6. Run the backend

cd ..
nat serve --config_file=backend/src/fair_repair/configs/config.yml

7. Backend is available at

http://localhost:8000

---

Usage

• Start the backend service.
• Open the frontend in a browser
• Submit a mechanic invoice (PDF or text).
• The AI agent analyzes cost fairness and outputs a structured transparency report.

---

Folder Structure

FairRepair/
├─ backend/
│  ├─ requirements.txt
│  ├─ src/fair_repair
│  │  ├─ configs/
│  │  │  └─ config.yml
│  │  ├─ __init__.py
│  │  ├─ main.py
│  │  └─ register.py
│  └─ pyproject.toml
├─ frontend/
│  └─ ...
├─ README.md
└─ LICENSE

---

Author

Cole Roberts

M.S. Artificial Intelligence, Santa Clara University

Email: cole03roberts@gmail.com

LinkedIn: https://www.linkedin.com/in/cole03roberts/

---

Acknowledgments

NVIDIA Agents for Impact Hackathon (2025) NVIDIA MCP, Agent Intelligence Toolkit, and NIM Nemotron.