Project Accomplishments: The Kinto pilot project explored how AI technologies—specifically sentiment analysis and generative language models—could enhance support for family caregivers of individuals living with Alzheimer’s and related dementias (ADRD). The team successfully extracted and anonymized 162 caregiver data sets using the Python tool “scrubadub,” after finding large language models (LLMs) unreliable for this task. These data sets included chat logs, coach notes, check-ins, and action plans. Using this data, they implemented an AI-driven caregiver strain assessment based on the REACH II protocol, which was validated against historical coach notes and deployed in Kinto’s production system. While the assessments were generally accurate and useful, some issues—like arithmetic errors and lack of sensitivity to recent life changes—highlighted areas for prompt refinement. 

In addition, the team developed a resource recommendation engine using a vector database of curated dementia care materials and GPT-4 to match resources to caregiver needs. The AI-generated recommendations were found to be relevant and helpful in nearly all test cases, though they often differed from coach-provided resources due to differences in context and timing. The project concluded that both the strain assessment and recommendation tools are feasible and valuable additions to the Kinto platform. Future improvements will focus on increasing responsiveness to recent caregiver activity and offering multiple tailored resource suggestions, with the goal of delivering timely, personalized support to caregivers. 

Initial Proposal Abstract: Numerous peer-reviewed studies show that caregiver support interventions that engage directly with caregivers to address the burdens of ADRD caregiving, result in significantly improved outcomes for both caregivers and Persons Living With Dementia (PLWD) with lower care costs. However, most traditional caregiver interventions are highly labor intensive and difficult to scale, particularly to underserved and remote communities.

Like traditional support interventions, Kinto relies on human coaches to create trusted relationships. However, Kinto leverages technology, including a mobile app, asynchronous text messaging, personalized content, and remote video, to amplify the reach of its coaches. We believe that AI can further amplify efficiency, and as an initial step we seek to apply AI algorithms to the caregiver and coach conversations that take place through the platform.

During this pilot project period, we will apply Sentiment Analysis on text messaging between coaches and caregivers to identify and prioritize elevated risk situations for which a proactive intervention may be warranted and on peer-to-peer support group text messaging to identify both negative situations which may need moderation as well as positive interactions that merit praise and confirmation. In addition, we will apply generative large language models to produce editable coach response templates that can reduce the time and effort required in composing responses to caregiver messages. Both efforts will lead to increased scalability of the Kinto platform by enabling coaches to support more caregivers simultaneously.

Outcomes:

• 
  Acquisition: Kinto is strategically acquired by Rippl CareApril 30, 2024
  SEATTLE, April 30, 2024 /PRNewswire/ — Today Rippl, a leading dementia-focused specialty care provider, announced it has acquired Kinto, a Boston-based, AI-enabled dementia caregiver support platform. This combination will fuel reaching an exponentially larger group of people living with dementia and their caregivers with human-centered care far beyond traditional bricks-and-mortar models. Kinto was founded… Read more: Acquisition: Kinto is strategically acquired by Rippl Care
• 
  Grant Funding: a2 Collective Pilot Award to explore LLMs as a support to caregivers of individuals with dementiaApril 19, 2024
  The Rippl team (formerly Kinto) led by Joseph Chung is part of a GY3 pilot award collaborating with McLean Hospital to evaluate Rippl’s LLM as a resource for caregivers of individuals with dementia and to determine if it gives accurate and useful information and appropriate emotional support to caregiver queries.… Read more: Grant Funding: a2 Collective Pilot Award to explore LLMs as a support to caregivers of individuals with dementia
• 
  New Product Launch: Addition of generative AI features to KintoFebruary 9, 2024
  The team at Kinto deployed the two main generative AI features to their production system where coaches can view strain assessment data and resource recommendations. The team also created a repeatable framework for extracting and anonymizing caregiver data sets from the Kinto platform on demand to provide training data sets… Read more: New Product Launch: Addition of generative AI features to Kinto