Here!

Problem & Origin

University student organizations and campus-wide assemblies regularly check in hundreds of students across gymnasiums, open courts, and lecture halls where cellular data is congested or unavailable. Manual sign-in sheets create long queues and transcription errors, while cloud-dependent biometric scanners stall whenever connectivity drops.

Here! began in 2023 as a college project built alongside Sheikha Encabo and Earlan Josh Sabillano. That original prototype proved that camera-based check-ins could replace paper rosters, but its architecture relied on sending captured frames to AWS Rekognition through Firebase Cloud Functions. Every check-in required a live internet round-trip and incurred per-invocation cloud vision costs—two constraints that made large-scale campus deployment impractical.

The 2026 Ground-Up Rebuild

In 2026, I re-architected Here! from scratch so the entire recognition pipeline runs 100% on-device inside a native Android application, paired with a static Astro web portal (youre-here.vercel.app) for public overview, privacy documentation, and APK distribution.

On-Device Vision & Embedding Pipeline

• Face Detection & Alignment: Uses Google ML Kit Face Detection directly on the device camera stream to locate faces, verify framing, and crop aligned regions at interactive frame rates.
• MobileFaceNet TFLite Embeddings: Replaces cloud vision APIs with an embedded MobileFaceNet TensorFlow Lite model that extracts normalized facial feature embeddings locally on the phone—reducing per-scan cloud inference cost to zero.
• Consent-First Face Data Lifecycle: Students explicitly consent before enrolling facial embeddings, and raw enrollment imagery is never exposed to unauthorized roles.

Offline-First Multi-Checker Sync

Campus events typically station multiple attendance checkers at different doors simultaneously, often without Wi-Fi.

• Deterministic Composite Keys ($uid~$timestamp): Attendance logs are written locally first via Cloud Firestore offline persistence using deterministic $uid~$timestamp composite identifiers, preventing key collisions when multiple checkers sync queued batches back to the cloud after an event.
• Role-Based Access Control (RBAC): Enforces distinct permissions for Students, Attendance Checkers, and Administrators across both the Flutter client and Firestore security rules.
• Turnout Velocity & Export: Organizers monitor real-time check-in velocity, late/absent breakdowns, and export verified CSV attendance sheets immediately after sync.

Outcome

Here! is currently live in Open Beta for Android (v1.0.0-beta.1), transforming what started as a cloud-tethered classroom prototype into a resilient, zero-API-cost field tool for student organizations.