marker-inc-korea / recipe-rag-demo

recipe RAG demo optimized by AutoRAG

recipe-RAG-demo

You are the developer of a recipe app. Your boss wants to create a new chatbot based on the recipe data. Ultimately, the chatbot can replace search engine in your app.

Now, you have to make the proof of concept for the chatbot in a few days. The only change to make it fast, yet powerful, is using AutoRAG. This is the journey to make the chatbot with the recipe.

Installation

pip install -r requirements.txt

Running the project

• Download dataset to data folder. You can download raw dataset from kaggle. Download RAW_recipes.csv only. That's all we need.
• Run preprocess.py to make corpus.parquet.
• Run make_qa.py and train_val_split.py to make qa.parquet.
• Make .env file using .env.template file.
• Run evaluator with the following command.

python main.py --config /path/to/config.yaml

• Check the result in the benchmark folder.

You can check the example config file at config folder.

And you can specify qa data path, corpus data path, and project dir if you want.