Vaka çalışması · Bitirme Projesi
Attendify
QR ve yüz tanıma tabanlı yoklama sistemi; 5 kişilik bir ekiple geliştirildi.
- Rol
- Backend sorumlusu
- Ekip
- 5 kişi
- Teknolojiler
- Python
- FastAPI
- PostgreSQL
- React
- Docker
- pytest
53
FastAPI uç noktası
13
PostgreSQL tablosu
47
pytest testi
3
React uygulaması
Bu vaka çalışmasının metni şimdilik yalnızca İngilizce.
Problem
Taking attendance by hand costs class time, and a signature sheet or a static code is easy to fake on behalf of someone who is not in the room. Attendify was our graduation project: an attendance system where a student checks in by scanning a QR code shown in class and confirming their identity with face recognition.
My role
I owned the backend in a team of five, and worked across the full stack wherever the API met the user interfaces. That meant designing the data model, writing the API, handling authentication and permissions, connecting the face recognition service, and making sure the three front-end apps could talk to the backend.
Architecture
The system has four parts:
- FastAPI backend: 53 endpoints over 13 PostgreSQL tables, with JWT auth, role-based access and Alembic migrations.
- PostgreSQL: the single source of truth for users, sessions and attendance records.
- Flask ML service: a separate service for face matching and liveness checks, called by the backend.
- Three React apps: one each for admins, instructors and students, all using the same API.
Challenges and decisions
One API, three kinds of users. Admins, instructors and students see very different data. Authentication uses JWT, and every endpoint checks the caller's role, so the same API can safely serve all three apps.
QR codes that expire. A QR code that never changes can be photographed and shared. The backend issues time-limited QR tokens for each session, so a code is only valid for a short window.
Keeping machine learning separate. Face matching and liveness checks run in their own Flask service instead of inside the API. The backend integrates with it as an external service, which keeps the main API focused on business logic.
Changing the schema safely. With 13 tables and a team of five, the database changed often. Every change went through an Alembic migration, so everyone's local database and the deployed one stayed in the same state.
Testing and deployment
I wrote 47 pytest tests covering the main flows: authentication, sessions, attendance and admin operations. The backend is containerized with Docker, so it runs the same way on every machine.
What I learned
Owning the backend of a team project taught me to think about the API as a contract: three apps and an ML service depended on it, so every change to an endpoint or a table had to be planned, migrated and tested. It also gave me practical experience with authentication, role-based access and integrating a separate service into a larger system.