- Software Engineering @ University of Novi Sad, Serbia
- Passionate about machine learning, computer vision, and mobile development
- Strong interest in algorithms, data structures, and practical software engineering
- GPA: 9.66 / 10 | Completed Harvard CS50 & CS50AI
🏋️ Android Barbell Tracker — Kotlin / TensorFlow Lite
On-device barbell detection and tracking for Android:
- Model: YOLOv8 nano optimized for mobile inference
- Tracking: SORT-based object tracking pipeline
- Features: Manual barbell selection, motion trajectory visualization
- Deployment: GPU-accelerated TensorFlow Lite inference on-device
🚕 FleetForge Taxi Platform — Spring Boot / Angular / Android
Full-stack Uber-like ride-hailing platform built as a university team project:
- Backend: Spring Boot REST API with JWT authentication and role-based access control
- Frontend: Angular web client with map-based ride tracking
- Mobile: Native Android app for passengers and drivers
- Features: ride assignment algorithms, pricing logic, driver availability constraints
🗄️ NoSQL Engine — C++
High-performance Key-Value storage engine:
- Architecture: LSM Trees, Write-Ahead Log (WAL)
- Optimizations: Bloom filters and probabilistic data structures
- Focus: scalable storage and efficient data retrieval
🧠 Word2Vec from Scratch — Python / NumPy
Implementation of the Word2Vec Skip-Gram model with Negative Sampling:
- Built entirely with NumPy (no ML frameworks)
- Includes training pipeline, negative sampling and evaluation
- Designed to deeply understand embedding learning mechanics
🧬 Genetic Algorithm – Black Box Optimization — Python
Applied evolutionary algorithms to optimize high-dimensional unknown functions:
- Real-valued GA with selection, crossover and mutation
- Population-based optimization and convergence analysis
- Graph Visualiser (Python): Modular plugin-based platform for visualizing graph data from multiple formats (JSON, CSV, XML)
- Traffic Computer Vision: CNN for German traffic sign classification with 95%+ accuracy
- Privacy–Utility Trade-off in AI: Studied how code obfuscation affects AI code completion models
- Nim AI: Reinforcement learning agent trained via Q-learning
- Checkers AI: Game agent using Minimax with alpha-beta pruning
- Parallel Scraper (C++ / Intel TBB): Multi-threaded scraping pipeline with task-based parallelism
- Minotaur Labyrinth Game (C++): Console-based game demonstrating OOP design principles
Languages: Python · Kotlin · C++ · Java · C#
Machine Learning: PyTorch · TensorFlow · TensorFlow Lite · scikit-learn · NumPy · NLP · Computer Vision
Mobile & Web: Android · Spring Boot · Angular
Other: Git · Algorithms & Data Structures · Parallel Programming · Software Architecture
