Kshitij Works

Independent AI products & engineering.
Gurugram, India.

← Engineering case studies

Android / Edge AI / Computer Vision

Rollingin

Rollingin is an offline-first Android attendance kiosk using entirely on-device facial recognition for employee enrollment and check-in/check-out.

Preparing for Google Play Role: independent builder & architect
Product overview & demo enquiries ↗
Recognition, on device
Rollingin logo
Offline face recognition attendance

The problem

A shared attendance kiosk needs to work when workplace connectivity is unreliable. Enrollment, recognition, and daily administration also need to fit the same application.

I built Rollingin, an Android tablet app covering employee enrollment, face-based check-in and check-out, attendance reports, scheduling, holidays, and administrator and manager access.

Recognition on the device

The interface uses React and TypeScript through Capacitor. Native Android components use CameraX for capture, MediaPipe for face processing, and LiteRT for on-device matching.

Enrollment guides the employee through multiple angles and checks face quality. The app stores face descriptors, pose-average and aggregate embeddings, and a small thumbnail for each face-profile version. It also retains recognition audit records. Full camera frames and enrollment video are not persisted.

Local application architecture

SQLite holds employee, attendance, and operator records. Authentication is local, with salted password hashes. The Android app sandbox protects the database; the database itself is not encrypted with SQLCipher.

Administrator and manager roles support workplace operations. Interfaces in English, Hindi, and 23 other languages, along with voice-guided scanning, are included.

Backup, sync, and sharing

Recognition is performed on-device, but the application supports administrator-initiated data transfer:

  • Optional HTTPS Cloud Sync and encrypted Google Drive/document-provider backups contain non-biometric business data, excluding operator accounts.
  • A full encrypted .faceid backup includes biometric records, attendance, reports, and hashed operator credentials. An administrator can copy it off the device.
  • Operators can share reports through email, WhatsApp, or Android’s share sheet.
  • Configured password recovery sends a recovery email address and one-time PIN through the organization’s SMTP server.

Data retention is part of the workflow

Deleting an employee removes them from active enrollment. It does not erase all their records: inactive face profiles, thumbnails, and historical recognition and attendance data may remain.

A factory reset erases the local application database, accounts, settings, and face-engine caches while preserving the device/site identity record. Exported files and external copies require separate deletion.

The app privacy policy and data removal guide explain these boundaries for administrators and employees.

Current status

The app is preparing for Google Play. Contact me for a demonstration or see the product overview.

Let’s talk

Need attendance or on-device recognition in a real workplace?

The same constraints — offline operation, on-device matching, kiosk workflow — show up in a lot of edge AI products.

Get in touch