MARI: Structured Decision Memory for LLM Recommendations
October 9, 2026
MARI improves recommendation performance in complex trade-off scenarios by replacing flattened token sequences with a Decision Memory Bank. This bank stores Structured Decision Memories containing user goals, constraints, and rationales to better handle similar items.
HOW THIS AFFECTS YOU
●
builderThis approach could improve recommendation engine precision for high-stakes consumer decisions.
●
researcherIt offers a method to move beyond sequence modeling toward explicit reasoning in recsys.