A very simple RAG implementation

This project is a simple RAG tool for asking questions related to some vnexpress articles.

This project is to demonstrate how RAG can be easily implemented without buzzy frameworks such as LangChain or LLamaIndex. Therefor, people can integrate RAG into their own system/project.

This projects uses:

• Scrapy for getting plain text of articles from vnexpress giao duc tin tuc
• Mistral Platform for both embedding and language models. They are mistral-embed and open-mistral-nemo
• Upstash Vector for vector database

Setup

Python 3.10

Install required packages, please see requirements.txt for extra information

pip install -r requirements.txt

Mistral API key and Upstash API key are stored at .env

MISTRAL_API_KEY=<key_here>
UPSTASH_VECTOR_REST_URL=<key_here>
UPSTASH_VECTOR_REST_TOKEN=<key_here>

Run

It is strongly advised to reach each .py file before running any command. By doing so, you get to understand the project more.

Scrap the data

At root project

scrapy runspider --set FEED_EXPORT_ENCODING=utf-8 src/vnexpress_spider.py -o data/articles.jsonl

Scrapy won't overwrite data/articles.jsonl if it already exists. If you want new data, you have to delete the file.

Setup the database

At root project

python src/setup_db.py

If a vector database already exists and data/articles.jsonl changes, you should delete the database.

Run the tool

You should definitely edit query variable in src/rag.py.

At root project

python src/rag.py

Great researcher/developer-friendly RAG frameworks

• stanfordnlp/dspy
• neuml/txtai