AI & Computer Vision Featured Case Study

Hoozoorino — Smart Biometric Attendance & HR Platform

Enterprise-grade smart attendance system powered by real-time face recognition, multi-frame anti-spoofing liveness detection, RBAC, and automated payroll reporting.

.NET 10 React 19 TypeScript Python FastAPI Face Recognition PostgreSQL TailwindCSS Docker

Live Demo View Source

Hoozoorino — Smart Biometric Attendance & HR Platform cover

Hoozoorino is a modern enterprise attendance platform built to eliminate physical hardware procurement bottlenecks and buddy-punching fraud. By pairing a sub-2.5s biometric computer vision pipeline with Clean Architecture on ASP.NET Core 10, it delivers touchless clock-in verification alongside comprehensive workforce management: custom shift definitions, leave approval workflows, immutable audit logging, and one-click payroll exports.

The Problem

Traditional biometric attendance systems (fingerprint scanners and RFID cards) carry heavy procurement costs, frequent hardware downtime, and persistent buddy-punching fraud. Basic webcam apps are easily spoofed with printed photos or smartphone replays, while HR teams waste dozens of hours each month manually reconciling timesheets and calculating overtime from fragmented spreadsheets.

The Solution

Hoozoorino resolves physical bottlenecks and HR operational friction through a centralized web kiosk: • Touchless Biometric Kiosk: Runs on any browser with an HD camera, verifying employee identity and vitality in under 2.5 seconds. • Role-Based Access Control (RBAC): Dedicated portals for System Administrators (shift scheduling, department hierarchies, leave approvals) and Employees (personal attendance records, leave submissions, early departure notes). • Shift & Holiday Management: Multi-shift definitions (fixed, rotational, floating), configurable grace periods, and official holiday calendar integration. • Leave Request Workflows: End-to-end employee leave requests (paid, medical, hourly) with real-time manager approvals. • Audit Logs & Compliance: Immutable audit trails tracking all administrative modifications and manual attendance adjustments. • Executive Dashboards & Excel Export: Live headcount analytics paired with multi-parameter Excel report generation for seamless payroll computation.

System Flow & Architecture

Built with Clean Architecture principles and decoupled microservices: - Frontend Layer (React 19 + TypeScript): Responsive kiosk UI with camera capture and instant employee feedback. - Core Business Layer (.NET 10 Web API): Shift computation rules, RBAC security, and PostgreSQL persistence. - Biometric Engine (Python FastAPI): Isolated InsightFace (buffalo_l on ONNX Runtime) computing facial embeddings in real time.

01 📷

Kiosk Camera

Webcam captures a 2-second live video stream upon employee arrival.

02 🛡️

Anti-Spoof Liveness

Multi-frame neural model analyzes blinking & motion to block printed photo/video fraud.

03 👤

Biometric Match

Facial vector embedding matched against encrypted database in <1s.

04 ⚙️

Shift Rules Engine

.NET Core evaluates active shift rules, grace periods, and registers entry/exit.

05 📊

Live Admin & Payroll

Instant headcount refresh on manager dashboard and one-click monthly Excel export.

Technical Challenges & Solutions

Multi-Factor Anti-Spoofing under 3s SLA

Challenge

Preventing fraud from static photos, tablet replays, and cutouts without creating entry bottleneck delays during morning employee check-in rushes.

Solution

Engineered a multi-frame pipeline analyzing 5+ consecutive frames. Combined Eye Aspect Ratio (EAR) for blink detection with 2D-to-3D landmark projection for horizontal Head Yaw rotation and facial tilt angles, executed in parallel using ONNX Runtime.

Impact: Zero false-positive photo bypasses with total inference latency kept under 2.4 seconds.

Biometric Privacy & Cosine Similarity Matching

Challenge

Raw employee photos cannot be permanently stored due to privacy compliance and database bloat. Verification must match against hundreds of registered staff with sub-second response times.

Solution

Extracted and encrypted 512-dimensional vector embeddings on initial enrollment. Attendance requests compute cosine distance against active employee embeddings, rejecting any score below the confidence threshold (0.65).

Impact: 100% compliance with biometric privacy norms and instantaneous matching for 1,000+ employees.

Automated CI/CD with SSL & Nginx State Preservation

Challenge

Automated GitHub Actions CI/CD workflows repeatedly overwrote the live Nginx reverse-proxy configuration on each push, discarding Let's Encrypt SSL certificates and causing service drops.

Solution

Decoupled infrastructure provisioning (Certbot SSL & Nginx configuration) from the rolling application deployment. Refactored the deployment runner to update binaries and trigger atomic systemd daemon reloads without touching Nginx site configuration.

Impact: 100% automated CI/CD pipeline with zero configuration drift and seamless zero-downtime updates.

Results & Metrics

• Sub-2.5s end-to-end verification latency from video capture to database commit. • 100% prevention of static image and replay-video spoofing attempts. • Automated payroll-ready shift reports, reducing administrative attendance processing time by 85%. • Comprehensive unit test coverage across Domain entities and Application business rules.

View Source Live Demo