Gradients Analysis

Created by:

• Diego Sabajo: https://www.linkedin.com/in/diego-sabajo
• Eitan Sprejer: https://www.linkedin.com/in/eitan-sprejer-574380204
• Matias Zabaljauregui: https://www.linkedin.com/in/zabaljauregui
• Oliver Morris: https://www.linkedin.com/in/olimoz

---

Overview

Explainable Diagnostic Assistant is an interactive tool designed to refine and analyze model outputs in a transparent and interpretable way. Built on top of Goodfire's API for advanced model evaluation and steering, this project focuses on improving trust and usability in AI systems, particularly in healthcare diagnostics.

Objectives:

• Find features that reduce hallucinations
• Build a medical hallucinations classifier that could also provide explainable results.
• Build a framework that demonstrates the use of feature steering to reduce the hallucination rate, while maintaining accuracy.

Technologies and Frameworks:

• Goodfire API:
• Python Libraries: NumPy, Pandas, Matplotlib.
• Streamlit:

Installation

Prerequisites

Ensure you have the following installed:

• Python 3.8 or higher
• pip (Python package installer)

Setting Up the Environment

1. Clone the repository

   cd project-folder
   git clone https://github.com/Mechanistic-Interpretability-Hackathon/Mech-Interp.git
   cd Mech-Interp
   

2. Create virtual environment

   python3 -m venv .venv
   

3. Activate virtual environment

   Mac: python3 -m venv .venv
   Windows: .venv\Scripts\activate
   

4. Install necessary packages

   pip install -r requirements.txt
   

Run the project

```bash
streamlit run src/app.py