Skip to main content
Back

Northeastern Rebuilds Vital Insight so Families Understand Aging Parents' Health

A new multi-agent LLM pipeline turns raw wearable data into narrative health updates for remote family caregivers, improving trust and satisfaction over raw data summaries

  • Researchers built a multi-layer, multi-agent LLM pipeline that turns wearable and smartphone data into narrative health summaries for remote family caregivers.
  • The system redesigns the open-source Vital Insight platform, shifting from time-ordered data dumps to insight-driven, context-aware narratives.
  • 11 remote family members were interviewed using initial summaries as technology probes; feedback drove a full system redesign.
  • The redesigned summaries showed significant improvements in satisfaction, helpfulness, trust, and willingness to receive updates compared to the original versions.
  • Three layered agents handle raw stats, behavioral interpretation, and personalized narrative -- each building on the last to answer "how" and "why,

Most remote family members caring for an aging parent get one of two things: a worried phone call or nothing at all. A new paper from Northeastern University researchers proposes a third option -- an LLM-generated narrative that doesn't just report what a sensor recorded, but explains how a loved one is doing and why. The work, titled "From 'What' to 'How' and 'Why'", is a significant step toward making passive health tracking actually useful for the people who care most about it.

The gap between data and understanding

Passive tracking methods like phone and wearable sensing have become dominant in monitoring human behaviors, but there remains a significant gap in translating these sensing streams into meaningful, high-level, context-aware insights required for real-world applications like summarizing an individual's daily routine. That gap is even wider when the audience isn't a clinician or researcher -- it's a son or daughter living across the country.

In the US alone, 63 million adults provide care to an adult family member or child with a medical condition or disability, and most of this care is unpaid. Nearly half of all adults age 50+ now use digital medical services, while 55% of unpaid caregivers age 50+ use one or more digital tools such as telemedicine, GPS/location detectors, and remote monitoring to coordinate care. The hardware is clearly there. The problem is making sense of what it collects.

Due to the nature of passive sensing data, sensemaking -- the process of interpreting and extracting insights -- requires both domain knowledge and technical expertise, creating barriers for different stakeholders. Existing systems designed to support sensemaking are not open-ended or cannot perform complex data triangulation. Remote family members (RFMs) are a particularly underserved group: they have deep personal knowledge of the older adult, strong emotional stakes, but limited time and no clinical training.

Starting with what already existed

The researchers didn't start from scratch. They leveraged an existing, open-source LLM system called