This work presents the design and implementation of an intelligent assistant embedded system that integrates emotion recognition with real-time student assessment on a low-cost hardware platform. The system is based on an ESP32 microcontroller with a camera module, responsible for capturing student facial expressions during test activities. A Convolutional Neural Network (CNN), pre-trained for facial emotion recognition, is deployed in a hybrid architecture where the ESP32 performs frame preprocessing and transmits data to a server running inference services optimized with TensorFlow Lite.

The approach is motivated by findings in affective computing and educational psychology, which emphasize the strong correlation between emotional states, concentration, and test anxiety. To estimate student readiness, the system employs a weighted mapping method that combines output probabilities with predefined emotion weights. For example, neutral or happy expressions contribute positively to concentration, while fear or anger increase nervousness. Normalized concentration and nervousness scores are computed over time, and readiness is determined using threshold-based rules: students are considered prepared when concentration reaches at least 60/100 while nervousness remains below 50/100.

An interactive dashboard allows teachers to monitor student states in real time and analyze historical performance. The proposed assistant demonstrates the feasibility of combining embedded devices, lightweight deep learning models, and real-time analytics for educational applications. By providing interpretable metrics of emotional conditions during assessments, the system offers a novel tool to support both student evaluation and teacher decision-making.