Research.

Two papers, one patent pending, and a pilot in 15 homes. Applied AI, tested on real people and real systems.

Cureus · August 2026

AI Caretaker: An Intelligent Voice-Activated Support System for Dementia Care

Aryan Dhingra, Charissa M. Nichols

I designed, built, and evaluated a tabletop device that talks with a person living with dementia, notices emergencies in ordinary speech, and logs meals, walks, and medication for the family. The study ran it through 120 scripted trials covering clear emergencies, safe controls, ambiguous cases, and long multi-turn conversations.

94.6%
speech-to-text accuracy
76.7%
emergency classification
120
evaluation trials
Read the paperdoi:10.7759/cureus.115149
AI Caretaker: An Intelligent Voice-Activated Support System for Dementia Care

IEEE Access, vol. 14 · October 2026

A Chain-of-Thought Prompt Engineering Framework for Automated Technical Specification Generation in Enterprise Integration: A Real-World Case Study

Aryan Dhingra, Gokul Menon, Anusha Antony

A six-stage, human-in-the-loop prompting framework for drafting the engineering specs a company signs off on before building an integration. Across 15 production interfaces at Metcash, authoring time fell from 793 to 300 person-hours. The gains shrank as interfaces got more complex, and most of the remaining human effort went into diagrams and cross-system logic.

62.2%
less authoring time
74.3%
content accepted unchanged
15
production interfaces
Read the paperdoi:10.1109/ACCESS.2026.3741017
A Chain-of-Thought Prompt Engineering Framework for Automated Technical Specification Generation in Enterprise Integration: A Real-World Case Study

Patent pending

AI Caretaker: A Voice-Activated In-Home Safety System for Dementia Patients

U.S. Provisional Application No. 64/073,388 · Filed May 23, 2026

Patent pending. Conversion to a full (non-provisional) application is underway.

Field work

Two-week family pilot

To get AI Caretaker out of the lab, I put devices in the homes of 15 Bergen County families caring for a relative with dementia. Real homes turned up problems that scripted tests never did.

Accents

Transcription accuracy dropped for some accents. The next version needs speech models tuned on more diverse voices.

Wi-Fi

Response time rose and fell with home Wi-Fi. That is the main reason I want to move the model onto the device itself.

Silence

Many patients went long stretches without speaking first. The device has to learn to start the conversation, not just answer.

Two hospital departments have also reviewed the project and asked to revisit it as the system matures.