https://github.com/langchain-ai/rag-from-scratch

RAG From Scratch

Query Translation

Process of taking a question and translating it into a form better for retrieval

Multi Query

Rag Fusion

Decomposition

Step-Back

• Create a more generic step-back question from the prompt

HyDE (Hypothetical Document Embeddings)

• Generate a hypothetical document from the query, and instead of using the embedding of the query, use the embedding of the generated document to retrieve relevant context.

Routing

Routing the potentially decomposed question to the relevant data source (DB)

Logical Routing

• Give knowledge of the various data sources and let the LLM decide which data source to use.

Semantic Routing