Classroom AI
A real-time, in-browser AI system that classifies student behaviour from facial expressions and head-pose estimation — designed to help Kenyan teachers gain objective insights into classroom engagement.
To provide Kenyan secondary school teachers with an affordable, privacy-respecting AI tool that offers real-time, objective data on student engagement and classroom behaviour — all running entirely within a standard web browser, with no external servers and no specialist hardware.
A future where every classroom in Kenya has access to AI-assisted teaching support — not to replace teacher judgment, but to augment it with data. We envision this technology reducing learning gaps caused by undetected disengagement and enabling more adaptive, responsive pedagogy.
Each detected student face is assigned one of 8 behaviour labels based on the dominant facial expression and head-pose estimation. Labels are colour-coded by severity.
Designed the AI detection pipeline, built the CSS design system and all front-end pages. Led research on facial expression mapping and classroom behaviour taxonomy.
Built the PHP backend, statistics module, and error logging system. Conducted field testing and wrote the research paper. Managed session data architecture.
Reviewed Paul Ekman's FACS framework, prior work on affective computing in education, and AI ethics in school settings. Identified face-api.js as the best fit for in-browser, privacy-preserving AI.
Built the AI engine, detection loop, behaviour classification mapping, and canvas overlay. Developed the CSS design system with dark mode, glassmorphism UI, and responsive layout.
Tested the system against classroom video recordings. Measured face detection rate (87%), behaviour accuracy (73% vs. manual annotation), and latency (400ms average).
Submitted to the 62nd Kenya Science and Engineering Fair in the AI & Technology category. Representing Maryhill Girls High School, Thika, Kenya.