Using RAG to build a movie recommendation agent
In the previous sections, we discussed what RAG is and how this system can be used to reduce hallucinations or extend model knowledge. As we mentioned, this system is composed of the following components:
- An LLM to generate the answer
- An encoder/retriever that transforms queries and documents into vectors
- A vector database where we save our vectors
We have, in this case, a dataset of movies and their description, and we want to create a system that, by asking a natural language question, will suggest the most suitable movies based on the information we’ve provided. Our LLM has no specific knowledge of the movies, and its parametric memory does not contain information about the latest releases. Therefore, RAG is a good system to supplement its knowledge.
- The first step is to obtain a corpus of chunks. Having taken a corpus of documents, we have to reduce it into chunks. A good compromise is to use...
