Compact Hierarchical Memory Engine (CHME): A Deterministic In-Memory Subsystem for Document-Grounded QA

Abstract

Compact Hierarchical Memory Engine (CHME) is a lightweight, deterministic, in-memory memory subsystem for document-grounded question answering. Rather than a monolithic RAG stack, CHME exposes stable APIs for ingestion, indexing, retrieval, prompt construction, LLM invocation, and snapshot persistence. The design emphasizes low operational complexity, reproducibility, and integration ergonomics for assistants, bots, and backend services. This paper specifies the system model, data structures, ingestion pipeline, retrieval and ranking logic, and persistence mechanics, and it summarizes computational characteristics and current limitations