Retrieval Augmented Generation

An approach to effectively create a “memory” for LLMs. The idea is to retrieve useful text from a database given user query just in time to augment the user query to generate the answer that makes use of the database.

Multiple RAG systems can be combined to simulate both short-term and long-term memories.

See LangChain

Overviews

• https://yang3kc.substack.com/p/how-to-build-an-effective-retrieval

Methods

Rerank

• Rerankers and Two-Stage Retrieval

Tools

• https://github.com/SciPhi-AI/R2R

Courses

• https://www.deeplearning.ai/short-courses/building-evaluating-advanced-rag/

Tutorials

• Say Hello to Precision: How Rerankers and Embeddings Boost Search
• “I want Llama3 to perform 10x with my private knowledge” - Local Agentic RAG w/ llama3