Implementation of RAG techniques to make code reviews in git repositories.

The project uses the following tools to work:

• LangChain
• Chroma as Vector DB
• GPT 3.5 Turbo as LLM
• ADA 002 for Text Embedding

To run the project first create a new Python environment and activate it. I'm using Anaconda for setting the python version that pipenv should use to set up the environment. The command bellow will automatically setup the environment with conda and pipenv:

make env

Now install all the project dependencies:

make install-all

Clone the repositories you want the model to code review:

make clone

You can change the repositories you want to clone inside bin/clone.py. Also take a look inside bin/insert.py when calling load_code_folder_chunks() function to verify which paths and code languages you want to include in the VectorDB. You can also define files that you want to exclude.

Insert the file chunks in Chroma database by running:

make insert

Ask something by running the following:

make ask

You can change your question inside bin/ask.py.