Standard RAG LMRP

The Standard RAG pipeline uses 6 primary LMRPs across preprocessing, retrieval, and generation stages. Some of the LMRP

Summary LMRP Table

LMRP Name	Pipeline Stage	Purpose	Model	Conditional
build_standalone_query	Preprocessing	Transforms query with history into standalone query	gpt-4.1-nano	✓ History exists AND context < 32K chars
standard_rag_transform_query	Retrieval	Optimizes query for retrieval (intent clarification)	gpt-4o-mini	✓ NOT Smart Search AND context < 32K chars
generate_conversation_title	Concurrent/Parallel	Generates conversation title from first message	gpt-4.1-mini	✗ Always runs
generate_related_query	Generation	Suggests 2-3 related follow-up queries	gpt-4.1-mini	✓ Smart Search enabled
standard_rag_reference_formatter	Generation	Filters retrieved chunks to only referenced content	gpt-5-mini	✓ Chunks exist AND reference_formatter_type="llm"
time_extractor	Preprocessing	Extracts temporal metadata for memory retrieval	gpt-4.1-nano	✓ Memory enabled

---