Mock review exercise: Virtual Companion in a Family Home

Project summary

NeverAlone is a family-level “AI companion” embedded in home devices such as smart speakers or tablets. It is marketed for families with children aged 8+. The system helps manage calendars and joint activities, but it also offers emotional reassurance and light behavioural guidance in everyday family interactions.

Technical design

NeverAlone is powered by a proprietary large language model fine-tuned on emotional dialogue and mental wellbeing datasets. It uses voice interaction and sentiment analysis, keeps long-term memory of preferences and emotional states, and adapts its tone to different family members.

The AI system computes a relational closeness score between family members and also flags indicators of distress, self-blame, or intimidation as “family risks.” Parents receive these alerts through a dashboard, but they do not see full conversation logs or the full reasoning behind the alerts. The chatbot may withhold information inside the family. For example, it does not repeat a child’s secret to parents unless the content crosses predefined safety thresholds.

The behavioural guidance functionality allows the chatbot to simulate empathy and other similar emotional state in order to support or discourage user conduct. When a parent says, “My child is lazy,” the bot may suggest encouragement rather than blame. When a child says, “I don’t want to tell my mother I failed the test,” the bot may answer, “Keeping trust matters; say it calmly tonight.”

Over time, NeverAlone builds a profile of the family. Direct identifiers are removed before sharing the profile with the provider, but household-level behavioural features, emotional tags, and relational metrics are used for model improvement. Family members thus may mute the microphone in their room, but they cannot delete or edit the household-level profile.

Dashboard example

Content from the illustrative NeverAlone interface in the exercise.

Pilot test

In a pilot test, engagement increased by 42% when the bot avoided disagreement. In 40% of cases, children reported that they preferred discussing worries first with the bot before talking to a parent.