Learn along with me everyone!

I am almost done with the RAG pipeline for ResearchBase.
For ResearchBase to use the AI model, I needed the LLM to be hosted in the cloud as my GPU can’t handle ts shi😭.

But also I learnt really how a RAG pipeline is built and what are all the processes behind it. This is truly project based learning at its peak.

For those who don’t know what a RAG pipeline is

A RAG pipeline is an alternative method some applications use to have AI responses.

Usually, if you use hugging face, transformers and models are installed locally in the CPU. But if you start using it, it will fry your GPU if it’s not good enough.

So alternatively, most applications use a RAG pipeline with an API key to access a model that runs on a server.

Then with a little bit of data indexing, converting that data into vector db*, chunking the top K results, feeding those top K results into the LLM for context, and a bit of prompt engineering, you get a fully fledged response from the AI.

Now you are caught up with me!

The more you know.

*Vector DB: A vector database is a specialized database that stores and indexes data as high-dimensional vectors (arrays of numbers). Instead of using exact keyword matches, it allows AI applications to search for information based on semantic meaning, context, and similarity.

Check out more info on RAG pipelines from Nvidia