Alexandru Gabriel Lapos

Motivated Software Developer with a strong technical background. I understand the core principles of how to write efficient code and solve complex problems. I can build applications using Python, PHP, Kotlin, and JavaScript. I am passionate about keeping my code organized and efficient, managing databases, and applying AI concepts to real-world projects.

Real Estate Management System (Project Voyager)

• Developed a full-stack web application using PHP and MySQL to manage real estate listings, supporting rental and sales transactions.
• Designed a relational database schema to handle complex data relationships between users (brokers/clients), property categories, and media assets.
• Integrated Google Maps API for dynamic location visualization and PHPMailer for system communications.
• Built a responsive frontend using Bootstrap 4, featuring dynamic image carousels and an intuitive property search interface.

PHP Authentication System

• Built a secure user management system using Object-Oriented PHP, implementing custom class autoloading for modular architecture.
• Developed a complete authentication flow including account activation via email and a secure password reset mechanism.
• Implemented session-based flash messaging to provide real-time user feedback and alerts using Bootstrap Toasts.

GPU Performance Analysis

• Analyzed the SGEMM dataset using Pandas and Seaborn to visualize correlations between matrix dimensions and computation time.
• Implemented data preprocessing workflows including feature scaling and train-test splitting to prepare datasets for predictive modeling.
• Evaluated model performance using Root Mean Squared Error (RMSE) to identify optimal configuration settings for GPU kernels.

BookSwap

A native Android application designed to facilitate the exchange of physical books between users.

Key Functionalities:

• Authentication: Users can create accounts and log in. This is managed via Firebase Authentication.
• Book Search: Users can search for books using an external Google Books API.
• Library Management: Users can save books to their personal library. Data is persisted using the Firebase Realtime Database.
• Meetings: Users can set up meetings to swap books. This feature integrates the Google Maps API to verify and display locations and tracks the status of the swap (Active vs. Rented).
• Dashboard: A central hub that displays user details, a horizontal list of library books, a vertical list of meetings, and navigation options.

The application is built using modern Android development standards:

• Language: Kotlin
• UI Framework: Jetpack Compose (using LazyColumn and LazyRow for lists)
• Architecture: MVVM (Model-View-ViewModel) to manage UI state and separate concerns.
• Dependency Injection: Hilt and Dagger.
• Backend: Firebase (Authentication, Realtime Database, and Firestore).
• Key Libraries: Retrofit (networking), Coil/Glide (image loading), Gson, Google Maps SDK

Comparative Sentiment Analysis: SVM vs. Deep Learning (LSTM)

Project Description:Designed and evaluated Natural Language Processing (NLP) pipelines to classify 50,000+ IMDB movie reviews as positive or negative. The project focused on comparing the efficiency of traditional Machine Learning (SVM) against Recurrent Neural Networks (Bidirectional LSTM) using custom-trained word embeddings.

Technical Stack:

• Language: Python
• Libraries: TensorFlow (Keras), Scikit-learn, Gensim (Word2Vec), NLTK, Pandas, NumPy, Seaborn/Matplotlib.

Key Technical Implementations

• Techniques: Text Preprocessing, Word Embeddings (Skip-gram), Cross-Validation, Transfer Learning.
• Developed an end-to-end NLP sentiment analysis pipeline achieving 86.4% accuracy, benchmarking Linear SVM against Bidirectional LSTM architectures.
• Trained custom Word2Vec embeddings (300-dim) using Gensim to improve semantic understanding of movie review data, handling a vocabulary of over 25,000 words.
• Implemented Stacked Bidirectional LSTMs with TensorFlow/Keras, utilizing Early Stopping and Dropout to mitigate overfitting during training.
• Conducted comprehensive Exploratory Data Analysis (EDA) and sentence length distribution analysis to optimize sequence padding strategies for Deep Learning models.