Capstone project (2025) · IoT × Mobile × Cloud
🚀 Developed as part of a 2-semester team course at FERI
This project was created as part of a 2-semester capstone course where the main challenge was to design an information system around the Direct4.me Smart Box.
The company provides an innovative technology for voice-token–based parcel delivery.
Our team expanded the idea into a self check-in/out system for Airbnb/Booking apartments, allowing guests to arrive at any time without manual key exchange.
Compared to existing lockbox solutions (owners send a code via WhatsApp), our approach enables better analytics and real-time insight into check-ins using the Smart Box technology as the core.
We were limited by the course requirements, which forced us to explore multiple technologies instead of choosing our own stack:
- Mobile app had to be built in Kotlin (Jetpack Compose) (not React Native).
- Face authentication had to be built from scratch (no external libraries).
- Everything had to run on a self-managed server with CI/CD pipelines.
This was intentional: the focus was on learning diverse tools and pushing into new areas, not just delivering a perfect product.
- Frontend: React 18, Tailwind, shadcn/ui
- Backend: NestJS (enterprise-grade Node.js framework)
- Microservice: FastAPI (Python) for face authentication
- Mobile: Kotlin (Jetpack Compose) mobile app
- Infra: Dockerized microservices, Nginx reverse proxy, Redis caching
- Database: PostgreSQL
- Object storage: MinIO (for images & AI models)
- Auth: Firebase OTP + custom 2FA (OTP or face recognition)
- Deployment: CI/CD pipelines → Amazon EC2 (free student tier)
- 📦 Parcel delivery integration with Direct4.me smart boxes
- 🏡 Self check-in/out for Airbnb & Booking hosts
- 📊 Dashboard for hosts (React web app)
- 📱 Mobile app for guests & cleaners (repo)
- 🔐 Custom face authentication (trained from scratch, Jupyter-based research → repo)
- 👩🔧 Cleaner support – dedicated mobile interface for housekeeping
- 🔄 CI/CD pipelines for automatic deployment on EC2
- 🗄️ Redis caching & MinIO storage
- 🧩 Modular monorepo setup (main repo)
A special part of the course required building face recognition without external libraries:
- We trained a custom model in Jupyter notebooks.
- Repo with research & training code: paketnik-face-auth-research-model.
- Due to GPU cost limits, the production service was excluded (fallback = always
true). - Live system uses Firebase OTP for real login security.
This project was less about AI and more about teamwork, architecture, and full-stack problem solving:
- Learned new tech under pressure (NestJS, FastAPI, MinIO).
- Understood the importance of pipelines, automation, and infra.
- Experienced cross-team coordination (frontend, backend, mobile, research).
- Delivered a solution that addresses a real problem (self check-in/out).
Even with limitations, the project pushed us into cutting-edge tools and gave us confidence to tackle enterprise-level development.
