Care Companion – Iman Azimi

What is Care Companion?

Care Companion is a proof-of-concept AI chat companion that helps family caregivers and patients with care challenges. It gets to know your situation, helps you find answers you can trust, and stays with you along the way. It answers questions about daily care, local services, and health coverage and benefits, and suggests actionable steps for the user.

Try the live demo

What problem does it solve?

Family caregivers carry a heavy load. They make hard decisions about someone else’s health, deal with confusing insurance and benefit paperwork, and have little time for themselves.
The help they need exists, but it is scattered, and the right information is hard to find. Programs and services differ from state to state, and their rules change often. Caregivers may search many websites to find one answer that fits their situation.
There is no solution that brings everything a caregiver needs together in one conversational package, fitted to each user’s needs. Existing solutions are generic and not tailored to the specific circumstances of the end users.

Why are general AI chatbots or solely skills-enabled assistants not enough?

Software is no longer the bottleneck, but trust is.

LLMs are emerging everywhere, revolutionizing how we interact with information. In addition, a growing number of new skills and capabilities are being added to AI chatbots. However, these advancements do not inherently guarantee trustworthy and reliable answers. Caregivers act on the answers they get, but an unreliable answer can cause real harm. A general AI chatbot may invent a phone number, quote last year’s rules, or give a generic answer that ignores the user’s state. To address trust, we need systems equipped with 1) mechanisms to verify the accuracy of information, 2) safeguards to ensure safety, 3) personalization approaches that tailor responses to individual users’ needs and contexts, among others.

How does Care Companion address these trust challenges?

This proof of concept aims to address these trust challenges by implementing mechanisms for safety, reliability, personalization, and actionability.

Safety: What is the risk or benefit of providing certain information or advice?

Care Companion first evaluates whether the input message might pose a risk to the user or to the system. If it detects potential harm, it either refrains from answering or directs the user to appropriate professional help, such as a doctor, nurse, pharmacist, or emergency services.

Reliability: What makes an answer reliable?

Care Companion bases its answers on trusted sources, not on the AI’s memory alone. Each LLM skill, a packaged set of instructions and tools for one task such as finding local services, is connected to specific, verified sources, ranging from reliable websites to local knowledge bases. This helps keep the information accurate and traceable to a credible origin.

Personalization: What makes an answer personalized?

Care Companion gathers contextual information over time, analyzes and connects the user’s preferences, needs, and past interactions, and builds a personalized understanding of the user. This allows it to tailor responses and recommendations to the individual’s unique situation, instead of providing generic advice.

Actionability and explainability: How does a response make a difference?

Care Companion bridges the gap between input requests and practical actions. It not only provides information but also 1) explains in plain language and 2) shows how to act on it via step-by-step guidance.

Current status, limitations, and future plans

Care Companion is at an early stage. It is currently a proof of concept in the testing and evaluation phase. The current version has limitations, particularly in the breadth of trusted sources and the depth of personalization. The next step is to seek feedback from end users and domain experts to reach acceptable performance.

What is Care Companion?

What problem does it solve?

Why are general AI chatbots or solely skills-enabled assistants not enough?

How does Care Companion address these trust challenges?

Current status, limitations, and future plans

Further reading