Retrieval Augmented Generation - Yet another acronym the ML researchers have slapped onto something so simple.
RAG WTH Is it?
Retrieval Augmented Generation - Yet another acronym the ML researchers have slapped onto something so simple. You have a large pre-trained model You have data lying in your documents, PDFs, CSV files How would you extend these models to infer insights from this data or in simpler terms, How can you make LLMs answer questions about your employee agreement your blog posts, or the data that is super private to your company? We can achieve it by performing 3 steps Retrieval Augmentation Generation Get it? Hence the name. Very innovative.
Ok…Ok.. wait…this is what every article in every major AI company is telling you. I can go two routes from here Explain to you how to do it. Again plenty of articles exist I highly recommend this video 🔗 Explain some cool stuff I learned in the form of prompt engineering Multi-query Given a question, 3 other questions of a similar nature are generated and consolidated result is sent back HyDE Takes LLM to a party. Given a question HyDE generates a hypothetical answer and then uses it to search the document you have Tokenization, and Vectorization are what I picked up during my chatbot days. I’m gonna leave them alone Oh wait! LLamaIndex has them all 🔗
Take Diversion into the World of Monitoring
it all started with the video and LangChain. People hate Langchain for it’s abstraction. But I loved the level of tracing it supports