The paper presents PIA, a memory and understanding agent for health applications. Unlike general that summarizes text, PIA extracts typed clinical records from conversations using a four-control harness: extraction (schema-constrained), memory (clinical gate, medical grounding), retrieval (decision-conditioned recall), and understanding (HU2). It shows that the same query gets different answers as memory depth increases from one-dimensional recall to a three-dimensional trajectory with causality.
Structured memory helps health agents interpret longitudinal patient conversations
PIA converts health conversations into time-resolved clinical records, then synthesizes them into evidence-backed user context.
Top University
KAIST · NAVER Corp.
Research Digest··2 min read
In one line
PIA turns health conversations into structured clinical records and synthesizes user understanding with four memory controls.
Before you read this
Three ideas this paper builds onagent memoryA repository where an agent stores past interactions and outcomes to inform future decisions.241 papers →knowledge graphA structured web of entities and relations, used to store domain facts and support reliable reasoning, retrieval, and model alignment.14 papers →memory retentionThe ability of a system to preserve and recall information over time or across contexts, often balancing compression against fidelity.32 papers →
Read it as
Yoon et al. built and deployed PIA as a memory and understanding layer for a consumer health agent.
The authors designed a memory harness with four controls: extraction, memory, retrieval, and understanding.
How it works
before and afterThe Daily Edition
The front page in your inbox each weekday morning: the lead, the picks, and why each one earned its place.