Query multiple documements using RAG! RAG is usually performed on a single document source or the entire knowledge base is treated as a single source. This RAG application can do much more than that!
The primary usecase this was built for was to query and compare 10K reports. An example query might look like this: "How much are Apple and AMD investing in R&D?". Now this question cannot be treated as a single question because information from both the AMD and Apple 10K report is required. This application accommodates for this.
- LangChain
- ElasticSearch VectorDB
- GPT3.5
Check out this notebook to get started. You'll find the end-to-end implementation of RAG here.
This folder has the spaghetti code written by me during experimentation and breaking my head while implementing this.
Check out the accompanying blog post over at datascience.fm.
Note: The packages installed in the blog post should suffice to run the notebook. I strongly reccomend you to not use my requirements.txt because it has a lot of uneccessary packages and you will end up downloading a lot more packages than necessary.