Analyse Retriever performance

[dasgoutam]

In a standard RAG pipeline, the first 2 steps usually involve 'Storage' and 'Retrieval' -

1. Storage - Load data from a source --> split data into chunks --> create embeddings --> Store in a data store(Vector DB)
2. Retrieval - On a given query, retrieve chunks from the data store

The performance of the retrieval process should depend on the following parameters -

• creating chunks in accordance to source data
• type of embedding model used
• retrieval mechanism used in the vector db

Using a single data source - 'muenchen-en', show results of how/whether performance varies on manipulating these parameters, and formulate an acceptable success criteria.