# Translation Memory In order to leverage translation memory effectively for in-context learning, we probably need to think through multiple mechanisms for memory retrieval. For example: 1. Selective memory: Agent can pull specific lines from data files and logs 2. Total abstracted memory: some kind of vectorized version of memory reduced with SVD or something along those lines to represent total memory in abstract form 3. Advanced RAG techniques ### Links to RAG resources 1. Academic survey of RAG techniques 2. Good chart of RAG techniques 3. Model each doc as structured metadata plus summary for multi-doc RAG 4. advanced RAG overview by founder of Llama index 5. Overview of RAG practice 6. No code RAG 7. RAG tutorial 8. Multimodal RAG “walking” a document 9. RAG plus agent 10. Application of an agent over a db 11. Use “this is the most relevant sentence in the context:” for better RAG 12. Rewrite user query for RAG with local LLM Interesting early discussion about [stateful bots](https://www.abrahamberg.com/blog/how-to-make-openai-stateful-text-generator-like-chatgpt-for-conversations/).