Intel & Predicition Guard Hackathon Challenge 5 Winning Project: Python Code Explanation Assistant

Introduction

Welcome to our hackathon-winning project! This repository contains the code for a Python code explanation assistant, developed by Simon Hanly-Jones, Ryan Mann, and Emmanuel Isaac, as part of the Prediction Guard and Intel hackathon. Our project leverages sequence transformers and Retrieval-Augmented Generation (RAG) to provide detailed explanations of Python code snippets.

Note that the resources made available during the hackathon have been de-provisioned and the API used as the foundation of the model is no longer authorised.

Overview

We aimed to create a Python code explanation assistant that provides accurate and detailed explanations of code snippets. Our approach involved two main models: one without RAG and one with RAG injection. By scraping official Python documentation and using the "all-MiniLM-L12-v2" sequence transformer, we enhanced the assistant's ability to retrieve relevant documentation when needed.

Models

Non-RAG Model

Our non-RAG model uses the following prompt template:

code_prompt_template = (
    "### Instruction:\\n"
    "You are a python code explanation assistant. Respond with a detailed explanation of the code snippet in the below input.\\n"
    "\\n"
    "### Input:\\n"
    "{query}\\n"
    "\\n"
    "### Response:\\n"
)

RAG Model

The RAG model uses this prompt template:

code_prompt_template = (
    "### Instruction:\\n"
    "You are a python code explanation assistant. Respond with a detailed explanation of the code snippet in the below input. Additional python documentation is to be used only if you do not understand the code snippet. \\n"
    "\\n"
    "### Input:\\n"
    "{query}\\n"
    "Python Documentation: {rag_context}\\n"
    "\\n"
    "### Response:\\n"
)

Key Findings

• Good prompting was crucial for performance. Referring to the RAG injection explicitly as "Python Documentation" helped obtain meaningful contributions from the RAG model.
• Using terms like “Context” or “Informational Context” confused the model, making it think the user was asking about the Python documentation instead of the code snippet.

Project Structure

Our code is split into two main files:

1. get_lance_db.py: Creates the database for RAG by transforming the pychunks.json file into a vectorized database.
2. wiz_code.py: Contains the chatbots and the FastAPI interface, and generates demo output to stdout before launching the FastAPI.

Running the Project

To run the project, follow these steps:

1. Download all files and place them in the same directory.
2. Run get_lance_db.py to create the vectorized database.
3. Run wiz_code.py to generate the demo output and launch the FastAPI.

The FastAPI will be available at 127.0.0.1:8000/docs, where you can access the automatically generated Swagger documentation as the user interface.

Demo Output

Example 1

Question:

print('Hello party animals!!!!!')

Answer with No RAG:

The code snippet is a simple print statement that prints the string "Hello party animals!!!!!" to the console. The exclamation marks are added to make the string more exciting and funny.

Answer with RAG:

The code snippet is a simple print statement that outputs the string "Hello party animals!!!!!" to the console. The Python documentation is not required for this explanation.

Example 2

Question:

x = max(2, 3)

Answer with No RAG:

The max() function returns the largest item in a given iterable (e.g., a list, tuple, set, etc.). In this case, the iterable is (2, 3). The max() function returns the largest item in this iterable, which is 3. Therefore, the value of x is 3.

Answer with RAG:

The code snippet is trying to find the maximum value of two numbers using the max() function. The max() function returns the largest item in an iterable or the largest of two or more arguments. In this case, the max() function is called with two arguments, 2 and 3. The max() function returns the largest of the two numbers, which is 3.

Team

• Simon Hanly-Jones
• Ryan
• Emmanuel Isaac

We are available for DnD-based coding challenges and promise to try not to break all provided hardware.

---

We hope you find this project as exciting and informative as we did! Feel free to reach out to us for any questions or further collaboration opportunities.