Semantic Search with history

[NajiAboo]

We had a requirement to list similar problems/issues based on the query entered by the User. It was working fine. But after ChatGPT, customers are asking similarly to ChatGPT. We need to remember the previous questions asked and have to find the solution based on the history. For example. In my first query If I asked something like " AC is not working in my Car" . as the first question. and in the second question If I asked " light is not turning on ", Then model has to relate it to the previous question and it should give priority to the Car and should bring the answers to solve the light issues in a Car ( at least give more priority /weightage to it) . How can we solve this kind of requirements using Haystack? any suggestion or tutorial pls

[silvanocerza]

Hey @NajiAboo, I think the best tutorial to follow for this use case is the Answering Multihop Questions with Agents .

As of now we're still working on creating a full fledged memory tool for Agents. Luckily the community came up with a basic memory library that does the job for now.

Hope this will be helpful. :)