Software engineer

RoCar - Car Price Predictor based on Machine Learning

This thesis introduces a production-grade web application, tailored for the Romanian automotive market, designed to predict vehicle prices. RoCar utilizes a model specialized in Romania's economy and pricing, trained on data we have independently scraped. The application stands out by using both structured data (year of production, manufacturer, model, options) and information from images and descriptions, leveraging a self-made multimodal architecture for more accurate predictions.

The Flinstones - Facial detection & recognition

Implemented a facial detection and recognition system for characters in "The Flinstones" using a sliding window approach and Convolutional Neural Networks (CNNs). The project involved patch extraction, binary and multi-class classification, and sliding window techniques.

Reinforcement Learning - Atari Skiing

Developed DQN and DDQN algorithms for OpenAI Gym Skiing environment. Showcased commitment to refining network architecture and preprocessing, addressing challenges in hyperparameter tuning. This project marked my initial venture into reinforcement learning implementations.

Double Double Dominoes

Implemented a computer vision system for automated dominoes game analysis. The project involves three main steps: board extraction, piece detection, and classification. Board extraction includes HSV masking, preprocessing, edge detection, and contour identification. Piece detection addresses challenges such as center text interference and misplacement. Classification utilizes templates for both vertical and horizontal domino pieces, employing template matching with preprocessing for accurate results. The final component, score calculation, involves dynamic programming considering player positions on the score board.

Brain Anomaly Detection

As a participant in a Kaggle competition hosted by my university's Artificial Intelligence course, I successfully created a binary classification model designed to identify brain anomalies within CT scans. This model helped me obtain the 8th place out of 128 students.

Calorie Tracker Backend

Developed the backend infrastructure for a React Native application, specializing in a calorie tracking feature similar to a simplified version of MyFitnessPal. Demonstrated a commitment to adhering to industry best practices in code quality and architecture. Notably, this project marked my inaugural endeavor in constructing a backend architecture from the ground up.