MAVN (MIPS Assembler of Higher Level) is a compiler that translates programs written in a higher-level MIPS 32-bit assembly language to basic assembly language. The compiler supports detection of lexical, syntactic, and semantic errors with appropriate error reporting. It implements register variables concept, allowing programmers to use variables instead of physical resources. The compiler performs liveness analysis, builds interference graphs, and handles resource allocation through graph coloring algorithms.

A machine learning project for predicting customer churn in banking using ensemble methods. Implements RandomForestClassifier, GradientBoostingClassifier, and AdaBoostClassifier achieving 80-88% accuracy. Features comprehensive data preprocessing including IQR-based anomaly detection, SMOTETomek for dataset balancing, and MinMaxScaler for normalization. Includes hyperparameter tuning with GridSearchCV, ROC curve analysis for model evaluation, and feature selection to identify key churn predictors.

Real-time control system for an automated package sorting facility developed in LabVIEW. Implements complex state machines for controlling conveyor belts, lifts, and sorting mechanisms across three floors. Features parallel processing for simultaneous package detection and system control, custom timer algorithms for conveyor belt management, Emergency stop functionality with state preservation, and SCADA interface for remote monitoring and control. The system handles package classification by size, automatic routing to designated floors, and real-time sensor-based decision making with comprehensive error handling and fault detection.

Full-stack fitness membership platform inspired by Fitpass.rs, built with Spring Boot backend and Angular frontend. Features comprehensive gym and fitness center management, membership subscription system with multiple tiers, facility booking and scheduling, user profiles with workout tracking, admin dashboard for managing facilities and memberships, and secure authentication with JWT tokens. Implements RESTful API architecture, PostgreSQL database for data persistence, and responsive design for seamless experience across devices.

Neural network-based stock price prediction system for NIS (Naftna Industrija Srbije) developed in MATLAB. Predicts minimum and maximum stock prices for the next five days using historical data up to May 2024. Implements multi-stage processing pipeline with parameter configuration, model training, and input calculation modules. Features time series analysis, pattern recognition neural networks, and financial data preprocessing. The system provides automated forecasting capabilities for investment decision support and market analysis.

Real-time license plate detection system using YOLO (You Only Look Once) object detection algorithm implemented in Python. Features automated license plate localization from images and video streams, high-speed detection with YOLO architecture, bounding box visualization for detected plates, and integration with OpenCV for image processing. The system provides accurate plate detection suitable for parking management, traffic monitoring, and automated vehicle identification applications.

Mobile application that uses device camera to capture photos and extract color information from them. Built with React Native for cross-platform compatibility. Features real-time color picking from camera-captured images, hex and RGB color code extraction, intuitive touch-based color selection interface, and color palette generation. Perfect for designers, artists, and anyone needing quick color identification on the go.

First freelance project on Upwork that directly connects device microphone to speaker with minimal latency. Built with React Native for mobile platforms. Features real-time audio routing from microphone to speaker, low-latency audio processing, adjustable volume controls, and simple one-tap activation. Developed as a karaoke and performance tool, enabling users to amplify their voice through their device speakers instantly.