I am working on a use case. I have an excel file which has 1000 text sentences. Each sentence is about a visit happened a dealer and discussion on product details. Now the user input will be question asked in general in the context of these 1000 text sentences. Once the user question is received, I have to summarize an answer from the 1000 datapoints. Now which would be the best approach to go ?
On the input processing side, I would probably go with a csv loader and use a RecursiveCharacterTextSplitter to chop it up. On the main chain, I would use a summarize chain and use “map-reduce” as a chain type instead of stuff:
from langchain.chains.summarize import load_summarize_chain
Alternately, you could use a LLMChain with prompt template to summarize the data and pass that as a tool for an agent.