If you want to be an AI app developer, then learn RAG – Handson 


So, you are part of a financial services organization and the organization decided that AI needs to be integrated into your organizations’ product and services, where do you start? Maybe, get an account to access one of the popular LLMs – Anthropic, OpenAI, Gemini or LLAMA. You start giving it some prompts and start to marvel at the responses.  

You show it your manager, and she types out a few queries. 

1. About the budget presented last year 

2. Asking about a press release that featured your company 

3. About your company’s stance related to unsecured loans that was posted last month 

Well, the answers were either outdated or plain wrong. 

The issue with querying an LLM with time sensitive information is that these LLMs will have no knowledge if it’s after the cutoff point when they were trained. 

Well, how do you fix this now? And make your manager happy with the responses? 

Enter RAG or Retrieval Augmented Generation. 

RAG is a means by which you provide additional context, additional data to an LLM so that questions like the ones your manager asked can be answered “correctly”, well almost.  

RAG is the best way for you to be able to make the LLM context specific to your product, application and organization.   

Well here are a few use cases you can solve for: 

1. Build a chat application for your customers with data about all the support tickets every logged and solved, release notes, quick starts etc. Users can get helpful hints that were given by the support engineers and also by other users in public forums. 

2. Build a better UI experience where instead of going through options on your Menu, user can give the intent and your “enhanced” UI understands the intent and brings up the input screen in one stroke 

3. A chat application for your organization instead of an elaborate intranet. You query and get to know all the policies, your leave information, and all the cool things you need at work 

4. A chat application that a doctor can query to get not only the latest case histories but also historical and research data 

5. And finally for the use case we started with, your manager queries your RAG enabled chat application, it not only gives historical information that was available when the LLM was last trained, but also the extra context that was added with RAG. 

To note is that in all these cases, a RAG pipeline adds the context of your application or organization to an LLM making it so much more useful to your organization.So, who do you provide this additional context with RAG? It’s by means of a RAG pipeline. 

A RAG Pipeline looks like this. 

  • A new question or prompt from a user -> Access relevant data from your data store-> Pass the relevant data + query to the LLM -> Receive the answer. 
  • And how do you prepare the data store, well there’s the data pipeline for it, and it looks like? 
  • Your raw data -> Chunk or tokenize the data -> Create embeddings -> Store this data in your datastore. 

Looks simple, right? 

Well, it does surely in theory. 

If you want to learn to do these steps hands-on, then join us for the session coming up on the 7th of June. 

Share the Post:

Related Posts