FairLens AI — Bias Detection & Mitigation Platform

FairLens AI is a hackathon-level MVP designed to detect, explain, and mitigate bias in datasets and machine learning models. Built for HR teams, banks, and healthcare analysts to ensure fairness and compliance.

📂 Folder Structure

FairLens/
│
├── backend/                  # FastAPI Backend
│   ├── main.py               # Core API & ML Fairness Module
│   └── requirements.txt      # Python dependencies
│
└── frontend/                 # React.js Frontend
    ├── public/
    ├── src/
    │   ├── App.jsx           # Main UI & State Management
    │   ├── index.css         # Tailwind CSS & custom styling
    │   └── main.jsx          # React Entry
    ├── package.json          # Node dependencies
    ├── tailwind.config.js    # Tailwind Config
    └── postcss.config.js     # PostCSS Config

🛠️ Setup Instructions

1. Backend Setup (FastAPI)

Open a terminal and navigate to the backend folder:

cd backend
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate
pip install -r requirements.txt
uvicorn main:app --reload

The backend will run on http://localhost:8000.

2. Frontend Setup (React)

Open another terminal and navigate to the frontend folder:

cd frontend
npm install
npm run dev

The frontend will run on http://localhost:5173.

---

🔬 ML Fairness Module (Core Logic)

The core ML logic is located in backend/main.py. It uses:

• Fairlearn: For calculating metrics like demographic_parity_difference and disparate_impact_ratio.
• CorrelationRemover: A pre-processing technique to remove the correlation between sensitive features and the rest of the dataset.

---

📡 Sample API Responses

POST /analyze

{
  "metrics": {
    "demographic_parity_difference": 0.21,
    "disparate_impact_ratio": 0.58,
    "bias_score": 58.0,
    "risk_level": "High"
  },
  "distribution": {
    "Male": { "count": 500, "favorable_count": 250, "approval_rate": 0.5 },
    "Female": { "count": 500, "favorable_count": 145, "approval_rate": 0.29 }
  },
  "explanation": "Analyzed outcomes across 'Gender'. Group 'Female' is 21.0% less likely to receive the favorable outcome compared to group 'Male'."
}

POST /mitigate

{
  "before": {
    "accuracy": 0.84,
    "demographic_parity_difference": 0.18
  },
  "after": {
    "accuracy": 0.82,
    "demographic_parity_difference": 0.04
  },
  "technique": "CorrelationRemover"
}

---

🚀 Deployment Guide

Backend on Render

1. Push your code to GitHub.
2. Go to Render and create a new Web Service.
3. Connect your repository.
4. Set the Root Directory to backend.
5. Build Command: pip install -r requirements.txt
6. Start Command: uvicorn main:app --host 0.0.0.0 --port 10000
7. Deploy.

Frontend on Vercel

1. Go to Vercel and create a new Project.
2. Import the GitHub repository.
3. Set the Root Directory to frontend.
4. Add an Environment Variable: VITE_API_URL pointing to your Render backend URL (if you change API_URL in App.jsx to import.meta.env.VITE_API_URL).
5. Click Deploy.

🎥 Demonstration Script

• Intro (0:00-0:15): "Meet FairLens AI, the platform ensuring fairness in your AI systems. Today we're analyzing a loan approval dataset."
• Detection (0:15-0:30): "We upload the data and instantly detect a 21% disparity in approvals between genders. FairLens explains this in plain English and flags it as High Risk."
• Mitigation (0:30-0:45): "With one click, we train a debiased model. The platform applies Correlation Removal, dropping the bias by 75% while maintaining 82% accuracy."
• Outro (0:45-1:00): "FairLens AI: Making AI fair, accountable, and transparent..."