Personal intelligence agent turns health talks into clear medical records
PIA: A Personal Intelligence Agent Turning Health Conversations into Records and Records into Understanding
Computation and Language
Summary
Conversations about health are often complex and hard for computers to turn into useful information. The authors created PIA, a smart helper that listens to health chats and makes clear, organized medical records from them. PIA can also understand a person's health journey over time, not just repeat past info. This helps answer tricky questions like changes in medication or blood sugar trends more accurately. The team also found that the way questions are asked can affect how well PIA understands the health information.
What this means in practice
- •For consumer health app developers: Integrate PIA to convert conversational health data into structured, meaningful medical records that improve user understanding and engagement.$Commercial implications: Enables consumer health apps to offer advanced personal health tracking and insight generation that can be marketed to end users.
- •For electronic health record teams: Use PIA’s domain-agnostic controls to enhance record accuracy by translating natural language conversations into typed clinical notes with causal context.
Authors
Jeonghun Yoon, Dongchan Kim, Hongyeon Yu, Young-Bum Kim, Jaegul Choo
Abstract
General-purpose agent memory summarizes conversations: it extracts salient snippets, embeds them, and retrieves the top-k into the prompt. A health agent cannot run on summaries: a dose becomes a sentence, "since last week" is resolved at the model's discretion, and a three-month glucose trend cannot be answered by text similarity. We present PIA, a personal intelligence agent deployed alongside a consumer health agent. PIA receives the agent's natural-language requests, decides for itself whether and how to write or read, and turns conversations into typed clinical records and records into a synthesized understanding of the user. Its memory harness consists of four controls -- extraction, memory, retrieval, and understanding -- each a domain-agnostic mechanism with a pluggable health module: schema, medical alias dictionary, knowledge graph, and temporal rules. We show how the same query receives a different answer as the memory injected into the response context deepens from one-dimensional recall, to a two-dimensional health snapshot, to a three-dimensional trajectory with causality, and report lessons from operation: self-reported health data are missing not at random, question phrasing governs the quality of synthesized understanding, and nearly a third of candidate causal links are structural noise that rules alone remove.