SnapAttend

AI based attendance management system

Comment

Inspiration

Attendance — one of the most tedious, time-consuming rituals in every classroom. Teachers call out names, students shout "present," and a good 5–10 minutes of every class evaporates into thin air. We asked ourselves: what if a single photo of the classroom could do all of that automatically? The idea of using facial recognition to snap attendance in seconds — literally just taking a photo and letting AI do the rest — is what sparked SnapAttend

What it does

SnapAttend is an AI-powered smart attendance system that automates classroom attendance using facial recognition and optional voice verification.

For Teachers:

  • Upload one or more photos of the classroom and SnapAttend automatically identifies which enrolled students are present
  • Manage multiple subjects, view per-session attendance records, and share subject enrollment codes with students
  • Alternatively, run voice-based attendance for audio verification

For Students:

  • Log in using just their face — no passwords, no IDs
  • New students can self-register by taking a photo and optionally recording a voice sample
  • View their enrolled subjects and personal attendance statistics in real time

How we built it

SnapAttend is built with a lean, modern Python stack:

  • Streamlit for the entire frontend — a two-portal web app (Teacher + Student) with session-state-based navigation and a clean responsive UI
  • face_recognition + dlib for facial embedding extraction and detection
  • scikit-learn (KNN/SVM classifier) trained on stored face embeddings to identify students from classroom photos
  • Resemblyzer for voice embedding generation, enabling optional voice-based attendance
  • Supabase (cloud PostgreSQL) as the backend — storing student profiles, face embeddings, voice embeddings, subject enrollments, and attendance logs
  • Pillow + NumPy for image preprocessing before feeding into the recognition pipeline
  • bcrypt for secure teacher password hashing
  • segno for QR code generation for subject sharing

Challenges we ran into

  • Face recognition reliability — getting accurate predictions from real-world classroom photos (varied lighting, angles, partial occlusions) required careful tuning of the face detection and classifier pipeline
  • Streamlit's stateless nature — Streamlit reruns the entire script on every interaction, which made managing multi-step flows (like the registration form staying visible after face scan) tricky. We solved this using persistent session_state flags
  • Audio input handling — st.audio_input returns a file-like object; using .read() depleted the stream before encoding, requiring a fix to .getvalue() to preserve the raw bytes for voice embedding
  • Multi-photo attendance merging — aggregating detections across multiple classroom photos without double-counting required building a per-student source map
  • Real-time classifier retraining — after each new student registers, the classifier needs to be retrained on the fly with train_classifier() without disrupting the app

Accomplishments that we're proud of

  • A fully working, end-to-end AI attendance system — from face detection to database logging — built and deployed as a real web app
  • Zero-friction student login: students authenticate with just their face, no credentials needed
  • Dual biometric support: both facial and voice recognition pipelines working in the same system
  • A clean, production-quality UI with subject management, enrollment flows, attendance history, and real-time stats — not just a demo
  • Self-registration for new students entirely within the app, including live face capture and voice enrollment

What we learned

  • How to build and serve a machine learning pipeline inside a Streamlit app without blocking the UI
  • The nuances of face embedding storage and nearest-neighbour classification at scale in a cloud database
  • How Streamlit's execution model differs from traditional web frameworks, and how to architect state-driven navigation around it
  • Integrating multiple biometric modalities (face + voice) into a single coherent identity pipeline
  • Using Supabase as a real-time backend with relational joins (e.g., subject_students → students) directly from a Python frontend

What's next for SnapAttend

  • QR code attendance — scan-in via a generated session QR code as a third verification method (infrastructure already started)
  • Mobile camera support — native mobile app or PWA so teachers can take attendance from their phone
  • Anti-spoofing — liveness detection to prevent photo-based spoofing attacks
  • Analytics dashboard — visualise attendance trends, at-risk students, and subject-level insights over time
  • LMS integrations — export attendance to Google Classroom, Moodle, or Excel with one click
  • Multi-face classroom accuracy improvements — better handling of occlusion, low light, and large class sizes using more advanced detection models (e.g. YOLOv8 face)

Built With

Share this project:

Updates

Submission history