Personalized ML Learning Agent

An Applied ML course project featuring an agent with context-aware memory, a reasoning planner, and adaptive inference stopping.

Features

• Adaptive Stopping: Simple questions bypass heavy reasoning to save tokens and time.
• Session Memory: Remembers your learning level and past topics.
• Gemini Integration: Uses Google's Gemini 1.5 Flash for high-speed ML explanations with low cost.
• RAG: Implements the slides from Applied Machine Learning and Applied Deep Learning as RAG and allows for uploading new slides.
• Secure Code Execution: Uses Docker sandboxing to safely run and verify Python code generated by the agent.
• Learning Dynamics: Record and Visualize Users' weaknesses in certain topics.
• Battle Mode and Interview Mode: Creative modes that test and challenge users' understandings.

Prerequisites

1. Python 3.9+
2. Docker Desktop (Required for the code execution feature)
   • Download and install Docker Desktop.
   • IMPORTANT: Ensure Docker is running in the background before starting the app.
3. Google Gemini API Key
   • You need a valid API key for Gemini 1.5 Flash.

Setup Instructions

1. Clone the Repository

git clone https://github.com/Sandman-R/AML-final-Project.git
cd AML-final-Project

2. Set up Virtual Environment (Recommended)

python -m venv venv
# Mac/Linux:
source venv/bin/activate
# Windows:
venv\Scripts\activate

3. Install Dependencies

pip install -r requirements.txt

4. Configure Environment Variables

Create a file named .env in the root directory and add your Google API key:

# Create .env file
touch .env

Content of .env:

GOOGLE_API_KEY=your_actual_api_key_here

5. Run the Application

You can start the server using uvicorn or directly via Python:

# Option 1 (Recommended)
uvicorn main:app --reload

# Option 2
python main.py

The server will start at: http://127.0.0.1:8000

How to Use

1. Open the UI: Navigate to http://127.0.0.1:8000 in your browser.
2. Upload Slides (Optional):
   • Click "Upload PDF" in the sidebar.
   • Upload your course slides (e.g., Lecture1_Intro.pdf).
   • Wait for the "Index Rebuilt" notification.
3. Ask Questions:
   • Try asking for code: "Write a PyTorch training loop." (Triggers Code Mode)
   • Try asking for math: "Derive the gradient for Log Loss." (Triggers Math Mode)
4. Battle Mode:
   • Click the "Battle Mode" toggle.
   • Propose a concept (e.g., "I think overfitting is good.") and see the agent challenge you.
5. Check Analytics:
   • Watch the Mastery Curve on the right update as you interact.
   • Click on any Weakness Tag to simulate an interview question for that topic.