AI Justice Assistant | Recidivism Risk Assessment

Project Overview

AI Justice Assistant is a machine learning-powered web application designed to assist legal professionals in assessing recidivism risk (the likelihood of a criminal defendant re-offending).

This project replicates and enhances the critical analysis of the COMPAS algorithm. It features a transparent, interpretable model that combines tabular defendant data with natural language processing (NLP) of charge descriptions to provide a holistic risk assessment.

Key Features

• Hybrid AI Model: Combines Logistic Regression (Tabular data) and TF-IDF/Naive Bayes (Text data) via a Meta-Learner.
• Glass-Box Design: Prioritizes interpretability to reduce "black box" dangers in judicial AI.
• Model Confidence Score: Displays the model's prediction probability to provide transparency on certainty.
• Bias Warning System: Automatically flags potential algorithmic bias for demographic groups with historically high error rates.
• Modern UI/UX: Features a responsive, glassmorphism interface for a seamless user experience.
• Ethical AI Focus: Built with fairness constraints and bias awareness as core principles.

---

Technology Stack

• Backend: Python, Flask
• Machine Learning: scikit-learn, joblib, pandas, numpy
• Frontend: HTML5, CSS3, Google Fonts
• Data: COMPAS Recidivism Racial Bias Dataset

---

Video Preview

AI.Justice.mp4

---

Installation & Setup

1. Clone the Repository

   git clone https://github.com/kidou16/AAIFinal.git
   cd AAIFinal

2. Create a Virtual Environment

   python -m venv .venv
   source .venv/bin/activate  # On Windows: .venv\Scripts\activate

3. Install Dependencies

   pip install -r requirements.txt

4. Run the Application

   python app.py

5. Access the App Open your browser and navigate to: http://127.0.0.1:5000

---

Project Structure

├── app.py                 # Main Flask Application
├── utils.py               # Model Logic & Utility Functions
├── models/                # Pre-trained ML Models
├── templates/             # HTML Frontend Templates
│   ├── index.html         # Input Form
│   └── result.html        # Prediction Result with Gauges
├── data/                  # Datasets used for training
├── requirements.txt       # Project Dependencies
└── notebooks/             # Original Analysis & Training Notebooks (.ipynb)

---

Disclaimer

This tool is a prototype for educational and research purposes. It is designed to demonstrate technical proficiency in Applied AI and ethical software design. It is not intended for real-world judicial decision-making without further rigorous validation and legal compliance checks.

---

Author: Yugal Jagtap