Page 89 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

(continued)

Item Details
Key Aspects of Solu- 1) AI Senior Care Content
tion 4.1-Healthcare
• Provides a wide range of structured programs tailored for seniors,
delivered through senior centres.
• Content genres include AI singing classes (e.g., Korean trot songs), AI
physical exercise sessions, and cognitive memory quizzes.
• Designed to enhance emotional well-being, physical health, and
cognitive stimulation.
• ESTsoft offers a solution called AI Studio Perso, a content creation
tool that generates AI avatars based on either real individuals (clones)
or custom personas. By inputting text, the AI avatar teacher delivers
spoken content. Using this solution, ESTsoft produces senior-targeted
video content featuring AI avatar instructors [4].
• The curriculum and scripts for these videos are collaboratively devel-
oped by senior healthcare specialists, university experts in brain
education, and certified instructors from the Korea Safety Education
Association [2].
2) AI Conversational Companion Service
• Offers real-time, interactive conversation with AI avatars to reduce
social isolation among homebound and single-living seniors.
• Integrates with smartwatches to track key health metrics, including
sleep quality, step count, and heart rate.
• In case of emergencies (e.g., abnormal heart rate or detected falls),
the system sends automated alerts to designated caregivers.
• Uses image-based food recognition and a health classification engine
to provide personalized nutrition analysis and dietary recommenda-
tions based on meal photos uploaded by the user.
• Delivered in Software Development Kit (SDK) form, the service is
compatible with various devices, including TVs, smartphones, tablets,
and PCs.

Technology AI Avatar, Large Language Model (LLM), Speech-to-Text (STT), Text-to-
Keywords Speech (TTS), Software Development Kit (SDK), Cognitive Training, AI
Companion Service, Content Recommendation, User Experience/User
Interface (UX/UI) for Seniors

Data Availability Private—collected via deployments in senior centres and partnerships
with medical/research institutions

Metadata (Type of text, audio, video, visuals, health metrics (nutrition in calories, sleep time
Data) in hours, walk in steps per minute, heartbeat in beats per minute), user
interaction logs
Model Training and The system leverages models specifically tailored for the senior care
Fine-Tuning domain. A custom-trained LLM is fine-tuned using real data from care-
giving scenarios to provide context-sensitive, emotionally attuned
conversational responses for both caregiving assistance and compan-
ionship.
In parallel, STT and TTS models are optimised for elderly speech charac-
teristics, such as slower pacing, pronunciation variation, and lower vocal
projection, ensuring reliable voice recognition and natural-sounding
responses. These models are further adapted using datasets that reflect
linguistic, behavioural, and cognitive patterns commonly observed in
older adults.

53