Demonstrates using a vector memory in a vector db qa chain using the chat-your-data challenge as a starting point. Rough idea is that instead of searching against a direct embedding of a question, the search is against a 'history vector' The history vector is formed by adding an embedding of the current question answer pair to a factor 'm' of the previous history For m < 1 this creates a vector that has some memory of the entire conversation by prioritises recent history
Create a ChatGPT like experience over your custom docs using LangChain.
See this blog post for a more detailed explanation.
Ingestion of data is done over the state_of_the_union.txt file.
Therefor, the only thing that is needed is to be done to ingest data is run python ingest_data.py
Custom prompts are used to ground the answers in the state of the union text file.
By running python app.py from the command line you can easily interact with your ChatGPT over your own data.