Hello dear community,
I am currently familiarizing myself with LLM, in particular RAG. I have an idea for a Q&A ChatBot and would like to learn how to build one.
I would like to have an Excel sheet with three columns as a data basis. The first column contains a typical question. In the second column, the question is provided with tags that describe the question. The third column contains an answer to the question.
When the user asks a question, a semantic search should first be carried out using the first and second columns. This finds the top of the most suitable questions in a cell. The answers to these questions in the third column are provided to the LLM as context.
It is important for me that the structured properties of the table are fully utilized. Conventional RAGs return a fixed chunk size, even if this spans cells. However, I only want the corresponding cell to be returned that relates to the question.
Which orchestrator is the best way to do this (LangChain, LlamaIndex,…)? Is it possible to return only the content of a cell? I have heard that LlamIndex is better for such data structures? Depending on which orchestrator I choose, I would familiarize myself with it first.
Best regards