InspirationWhat inspired us
Attendance in most institutions is still manual, proxy-prone, and hard to audit. We wanted to build something that feels modern, trustworthy, and practical for real campus workflows — not just check-in, but prevention, analytics, escalation, and action.
What we built VerifAI is an AI-powered attendance and monitoring platform that combines:
Face scan with anti-proxy checks (including blink verification and liveness signals) QR session attendance for fallback and low-friction check-ins Student risk tracking (below 75%), appeals workflow, and teacher review Events and exams modules (identity checks + integrity/proctoring workflows) Real-time notification pipelines (WhatsApp + Telegram text) AI admin copilot chatbot for attendance-focused operational insights How we built it We used a modular architecture:
React + Vite frontend for fast iteration and responsive UX face-api.js stack for browser-side recognition, landmarks, and blink gating Node/Express backend for external notification integrations FastAPI backend for service endpoints and extensibility localStorage/sessionStorage for rapid prototyping and portable demos Groq-powered admin chatbot with strict attendance-only system prompts and live local data context Challenges we faced
Balancing detection speed vs recognition reliability in real-world lighting/device conditions Reducing false positives and unknown-loop behavior during continuous scan streams Hardening anti-proxy logic without degrading user experience Making notification integrations reliable across local, LAN, and hosted environments Handling push-protection/secret hygiene while moving fast Keeping a growing feature set coherent across admin, teacher, and student roles What we learned
Liveness is not one signal — combining motion, landmarks, and blink improves trust significantly Operational products need workflow depth (appeals, logs, reports), not just model outputs UX decisions (mobile-first layouts, fixed nav patterns, response clarity) impact adoption as much as AI quality Prompt-constrained AI assistants are most useful when grounded in live system data and strict domain scope Reliability, observability, and secure config management matter from day one — even in hackathon builds Why this matters VerifAI turns attendance from a passive record into an active intelligence layer for institutions: reducing fraud, surfacing risk early, and enabling faster data-driven academic decisions.
Built With
- callmebot-telegram-text-api-exports-&-reporting:-jspdf-storage/db:-browser-localstorage-(app-data)
- css
- eslint
- fastapi-(python-apis/mock-services)-notifications/apis:-twilio-whatsapp-api
- framer-motion
- git/github-deployment-target:-vercel-(frontend)-+-separate-backend-hosting-(e.g.
- groq-api-(llama-3.3-70b-versatile)-for-admin-chatbot-qr-stack:-html5-qrcode
- html-frontend:-react-(vite)
- languages:-javascript
- postcss/autoprefixer
- python
- qrcode-backend-services:-node.js-+-express-(notifications/integrations)
- react-qr-code
- react-router
- recharts-ai/computer-vision:-@vladmandic/face-api-(face-recognition-+-landmarks-+-blink/liveness)
- sessionstorage-(chat-session-history)-tooling:-npm
- tailwind-css