educational apps measure scores and completion rates — but they never capture how a child actually feels while learning. a child can get every answer right while silently frustrated, or fail every question while genuinely engaged. that gap makes it hard for parents to support their child's real learning experience.

emoshield integrates real-time facial emotion recognition directly into iq-based cognitive games for children aged 6–10. the app captures the child's emotional state during play — not just after — and builds a report parents can review. games adapt difficulty based on performance, and the emotion data is stored per session so parents can track patterns over time.

ran inference on-device (not cloud) for faster predictions, offline capability, and better privacy — children's facial data never leaves the device. trained the model in python with tensorflow and converted to tflite for mobile deployment. mediapipe handles face detection before the emotion model runs, keeping prediction noise low. confidence threshold set at 0.45 — anything below is discarded rather than showing a wrong emotion.

gives parents visibility into their child's emotional engagement during learning — not just test scores. emotion trends across sessions let parents and teachers spot patterns: does the child feel angry during timed tasks? does fear spike before hard levels? this is the data that score-only apps miss entirely.