To set up basic vector database RAG via Vext, you need to perform three tasks:
Add a Data Set and a Data Source
Add a "Vector Database" action
Add an "LLM" action
Add a Data Set and a Data Source
The most important piece of an RAG (retrieval augmented generation) is to import the data/file. Check out here to learn more.
Add a "Vector Database" action to the AI Project
Once you've added a Data Set and at least one Data Source, you can now create (or select) an AI project and add a "Vector Database" action to refer to the Data Set. Check out here to learn more.
Add an "LLM" action to the AI Project
You will need to add an LLM behind the "Vector Database" action to generate meaningful response based on the retrieved data. To correctly refer to the output retrieved from the Vector Database action, check out here to learn more about variables.
Bonus: Why Adding LLM behind the "Vector Database" action?
Vext "AI Project" works like a chain, each action's output will be available as a variable and can be utilized in a prompt within an action afterwards. So for a standard Vector Database RAG, it looks like:
So it's critical to put the LLM behind the data retrieval action so it can utilize the retrieved data for the question.
Here's an example of how the LLM system prompt looks like after the vector database action (for Llama 3.1):