Golden Retriever

chatting with the model
all matching items in the query

Inspiration

We wanted to create a more customizable RAG pipeline. One of the common issues with RAG is the chunk breaking in the middle of important information. We wanted to separate the retrieval and generation, so that the user can see the retrieved chunks in the context of the original document. For the generation step, we wanted to allow the user to reposition or resize the chunk window as well.

What it does

The chunks are embedded with openAI API and then match the chunks to the original text to display. LanceDb is utilized to search for those chunks and retrieve them. User is prompted to put in an input that is a prompt for searching, then the program should compile the context and send it using OpenAI to generate a response.