Is there an existing name for this RAG-like technique I've come up with?

SUMMARY: It's a form of chat memory, to enable recall of difficult prompt-answer pairs.

OVERVIEW: Anytime the LLM struggles or fails to generate a correct answer on first try, the final corrected answer is stored in a vector key-value database, with the prompt as the key and the answer as the value. In the future, if a very similar prompt is presented, past answers are injected into the prompt as n-shot examples.

EXISTING: This technique seems obvious but I haven't …

aipromptprogramming chat database form generate key llm memory overview prompt rag recall summary the key the prompt value vector