In this module, we will talk about how to increase accuracy of RAGs. Big companies require their data outcome from LLM’s to be accurate, so there are many techniques developed for the same. First, let us see what is difference between Basic RAG and Advanced RAG. While basic RAG has only 3 steps , Advanced RAG includes 5 steps:

1. Query Transformation

2. Routing

3. Query construction

4. Indexing

5. Retrieval

6. Generation

Query Transformation

The main aim of query transformation is to convert the user prompt from what user wrote to what user actually want. It deals with the concept of abstraction.(abstraction meaning)

Sometimes, we require query given by user to be more abstract or less abstract. Giving more abstraction or less abstraction completely depends on user query and type of method we use for Query transformation and the use case of RAG model.

In the upcoming module I will list all the methods of Query transformation and code for the same. Stay tuned!

Advanced methods for Query Transformation

Link for Parallel Query Retrieval: Parallel-Prompting

Link for Reciprocal Rank fusion: Reciprocal rank fusion

Link for Query Decomposition: Query Decomposition

Link for HyDE and Step Back Prompting: https://shreypaunwala.hashnode.dev/hyde-and-step-back-prompting