ML Exam Generator

This project is a multi-agent system designed to automatically generate exams based on provided educational materials. It utilizes autogen-agentchat for agent orchestration and chonkie for semantic document processing.

Overview

The system works by:

1. Ingesting Content: Reading markdown files from the docs/ directory.
2. Indexing: Chunking the content using semantic analysis and storing it in a ChromaDB vector database for efficient retrieval.
3. Generating Exams: Using a team of AI agents (Planner, Creator, Verifier, Reviewer, Saver) to collaborate on creating, verifying, and saving exam questions.

Prerequisites

• Python 3.12+
• Google Gemini API Keys

Installation

Option 1: Using uv (Recommended)

1. Clone the repository.
2. Install the dependencies using uv:

   uv sync

Option 2: Using pip

If you prefer using standard pip:

1. Install dependencies:

   pip install -r requirements.txt

Configuration

1. Create a .env file in the root directory.

2. Add your Gemini API keys. You can rotate between multiple keys if needed:

   GEMINI_EXAMK1=your_api_key_here
   GEMINI_EXAMK2=your_second_key_here
   GEMINI_EXAMK3=your_third_key_here

Usage

1. Indexing Content

Before generating exams, you need to index your source materials. The src/chunker.py script reads markdown files from the docs/ folder, splits them into semantic chunks, and stores them in the local ChromaDB instance.

# Using uv
uv run src/chunker.py

# Using python directly
python src/chunker.py

2. Running the Generator

To start the exam generation process, adjust the topic variable in main.py and run the main script. This initializes the agent team and starts the workflow. The agents will collaborate to create an exam, which will be saved in the exams/ directory.

# Using uv
uv run main.py

# Using python directly
python main.py

Project Structure

• docs/: Source markdown files for the exam content.
• exams/: Generated exam files.
• src/: Source code.
  • agents/: Definitions for the different agents (Planner, Creator, etc.).
  • models/: Configuration for LLM clients (Gemini, Ollama).
  • chunker.py: Script for indexing documents.
  • memory.py: Vector database interface.
• chroma_db/: Local storage for the vector database.
• .env: Environment variables for API keys.