This study presents a novel robotic assistant tailored for elderly rehabilitation. The system combines state-of-the-art artificial intelligence, computer vision, and robotics to provide customized exercise routines and instant feedback. The core of the assistant is a Grove AI V2 vision system, driven by a dual-core Arm Cortex-M55 processor and an Arm Ethos-U55 neural network component. A custom-built robotic torso fitted with 12 servo motors supports the demonstration of a wide variety of upper body exercises. A crucial element in the experiments is the use of Restricted Radial Equivalence Functions (RREFs) and Radial Similarity Measures (RSMs) to precisely match human poses with the robot’s demonstrations. This method ensures an accurate evaluation of exercise performance, enhancing the efficacy of physical therapy for the elderly. Using edge computing technologies, the system provides real-time monitoring and personalised care while addressing important privacy issues. The assistant is additionally characterized by a small-sized design that includes a 124 \(\,\times \,\) 64 pixel OLED display, capable of displaying bio-signal monitoring and an animated face to engage users. This research advances AI-assisted elderly care, aiming to increase independence, physical health, and overall quality of life for aging populations while offering support to healthcare systems and caregivers.