AWS re:Invent 2025 - Detecting falls in aged care with a minimum lovable product (DEV337)

Detecting Falls in Aged Care with a Minimum Lovable Product

Overview

• The presenter shares a personal story about their grandmother's fall in an independent living facility, which inspired them to start a company addressing this problem.
• The presentation covers the problem being solved, requirements gathering, minimum viable product (MVP) vs. minimum lovable product (MLP), data enrichment, real-time escalation, and key takeaways.

The Problem

• Typical aged care rooms use outdated technology like pressure floor mats, bed alarms, and infrared switches to detect falls and movement.
• These technologies are expensive, can be unreliable, and don't provide real-time alerts when someone falls and needs immediate assistance.

Proposed Solution

• The presenter's startup uses a millimeter wave radar device installed in the ceiling of an aged care room.
• The radar device can detect the X, Y, and Z coordinates of people in the room, identify if someone has fallen, and send real-time alerts.
• This solution aims to provide more reliable and actionable fall detection compared to the existing technologies in aged care facilities.

Personas and Requirements

• The two key personas are:
  1. Aged care facility managers, who need reporting, prioritization of care, and compliance recording of falls.
  2. Nurses, who need real-time alerts when someone falls to provide immediate assistance.
• The solution needs to address both batch reporting requirements and real-time alerting needs.

Minimum Viable Product (MVP)

• The initial MVP was a Grafana dashboard connected to AWS services like Athena and CloudWatch.
• This allowed the presenter to quickly demonstrate their understanding of the data and query capabilities, even though it was a single-tenant solution.

Minimum Lovable Product (MLP)

• To address the limitations of the initial MVP, the presenter added the following enhancements:
  • Tenant context enrichment to the event payload
  • Separate data partitioning and storage per tenant using Amazon S3 and Athena
  • Real-time event processing and escalation using Amazon SNS, AWS Lambda, and AWS Step Functions

Real-Time Escalation

• The solution includes a real-time escalation workflow using AWS Step Functions:
  1. Fall event is detected and sent to an SNS topic
  2. An AWS Lambda function processes the event and stores it in a DynamoDB table
  3. An EventBridge pipe triggers an AWS Step Functions workflow
  4. The Step Functions workflow has multiple levels of escalation with delays, checking for acknowledgment at each level

Key Takeaways

• Start with a minimum viable product (MVP) to prove the concept works, then build a minimum lovable product (MLP) based on user feedback.
• Utilize low-code/no-code tools like Grafana to quickly build initial prototypes and dashboards.
• Enrich event payloads with metadata like timestamps and tracing information to aid debugging and analysis.
• Understand the balance between batch and real-time processing needs - only add complexity for the real-time requirements.
• Visualize events and data to provide valuable insights to end-users.
• Leverage AWS services like IoT Core, Athena, DynamoDB, and Step Functions to build a scalable, serverless solution.

Business Impact

• The presented solution aims to provide aged care facilities with a more reliable and responsive fall detection system compared to the outdated technologies currently in use.
• By enabling real-time alerts and escalation, the solution can help ensure that elderly residents receive immediate assistance after a fall, potentially improving health outcomes and reducing the severity of injuries.
• The data collection and reporting capabilities can also assist aged care facility managers with compliance, resource allocation, and understanding resident movement patterns.

Example Use Case

• The presenter demonstrated a use case where a resident falls out of bed, triggering a real-time alert that is escalated through the Step Functions workflow until acknowledged by the appropriate caregiver.
• The solution's ability to detect the resident's location, center of mass, and fall event allows for a targeted and timely response, rather than relying on the resident to manually activate a call button.