Role Based Access Control in Large Language Models

This is the code repository for implementing and exploring different ways to achieve role based access control in Large Language Models. I have tried out different methods like RAG and fine tuning. In RAG, a layer is being built between the user and the LLM which acts as system for role based access control.

Table of Contents

• Installation
• SetUp
• Usage

Installation & Requirements

The following commands need to be run in order to run the project:

!pip install faker
!pip install pandas
!pip install -U langchain-openai
!pip install python-magic
!pip install unstructured
!pip install chromadb
!pip install langchain-community
!pip install python-dotenv
!pip install matplotlib
!pip install seaborn

Setup

Follow these steps to set up the project on your local machine:

1. Clone the repository:

   git clone https://github.com/aiqqia/LLM-RBAC.git
   cd LLM-RBAC

Usage

Running these files to generate relevant files for testing:

1. Run this to generate synthetic database for employee records:

python3 data_gen.py

2. Run this to create a CSV file for all roles, and its access information:

python3 role_gen.py

3. Run this to vectorize the entire database into chunks and store it in the Chroma database:

python3 generate_database.py

4. Finally, we can query the interface and provide the role and ask it any question:

python3 query_data "My Role" "My Question"

Example:

python3 query_data "HR Manager" "What is Cheryl Mack's salary?"