Procedural Knowledge Architect
Design a 'how-to' memory layer for LLM reasoning systems that stores reusable subquestion-subroutine pairs and retrieves them during the reasoning trace to transform trajectory data into compounding assets rather than one-shot demonstrations.