I have been working on my latest chat iteration and found that if I leave the initial input_text alone I either get no response (30%) or a very simple answer from my vectored list of question / answers (70%).
I decided to combine the intermediate response with my initial input and have been super happy with the detail and cross-pollination of answers I am receiving. I am now getting about 10% no response and more detailed answers the remaining 90%.
search_results = vectorstore.similarity_search(vector_text, k=1)
intermediate_page_content = search_results[0].page_content
# Combine intermediate_page_content with the original input_text
combined_text = f"{vector_text}? {intermediate_page_content}"
Is this common? Are there better ways to do this? What challenges am I going to face with this?
quick side note, by leaving the question in there I did get one response saying for your first question x and the second question y. Which is ok but the end user did not ask two questions so that could be confusing.
Love to hear your thoughts.