Secure Visual Classification Platform
Bachelor's Thesis. Grade: 10/10
A secure full-stack platform architected around its trust boundaries. A Flask REST API enforcing authentication and role-based access control sits over a multi-database design — PostgreSQL with parameterized queries, Redis for secure session state, and MongoDB for isolated log layers — with endpoints validated per OWASP and Docker for runtime isolation. Visual classification and preference inference are built on top of that secure foundation.
Project Details / Background
Architected the platform around security from the start: a Flask REST API enforcing authentication and role-based access control, with Redis managing session state. Input validation and endpoint hardening followed OWASP guidance, and the system ran in Docker containers to isolate its runtime.
Designed a multi-database persistence layer where each store has a clear security boundary — PostgreSQL with parameterized queries for relational data to close off injection, Redis for secure session and cache state, and MongoDB for isolated log and telemetry layers. Separating these concerns keeps data access patterns clean and limits what any single component can reach.
On top of that secure foundation, I built the computer vision pipelines and AI models for multi-label image classification and attribute extraction. The models were trained with PyTorch and integrated into the Flask backend for real-time inference.
Implemented inference logic for preference modeling based on historical interaction data and visual feature embeddings, so the system learns user preferences from how they interact with classified content.
Applied Scrum methodology adapted to a solo developer workflow with iterative planning and continuous integration, ensuring structured development and consistent delivery throughout the thesis project.
Technologies
Flask • REST APIs • Authentication & RBAC • OWASP • PostgreSQL • Redis • MongoDB • Docker • Flutter • PyTorch • Computer Vision • Scrum