Hi, I'm Sara Etesami

Fine-tuned BERT-base-uncased on AG News and SST-2 datasets for text classification. Built an end-to-end training and inference pipeline using Hugging Face Transformers and PyTorch, achieving ~93% accuracy and 0.91 weighted F1-score. Optimized hyperparameters and batch size, cutting training time by 25%. Supports both news category classification (4 classes) and sentiment analysis.

Built a versatile LSTM-based time series forecasting pipeline in Python, designed for energy load prediction and adaptable to any continuous signal (weather, sensors, industrial trends). Implemented end-to-end: data windowing, feature engineering (calendar + weather features), MinMax scaling, rolling-origin validation, multi-step decoding, and a backtesting framework. Achieved ~0.35°C next-hour MAE with consistent generalization across 5 rolling folds. Packaged with CLI tools for training, evaluation, inference, ablation studies, and forecast visualization.

EN32-Extended-GOL is an enhanced implementation of Conway’s Game of Life built in Java, expanding the classic cellular automaton with advanced game mechanics and persistence. The project introduces specialized cell types such as Highlander, Loner, and Social, along with mood-based behaviors like Naive, Healer, and Vampire that enable energy transfer and transformation. It also features an interactive board with energy-modifying tiles, global events that dynamically affect gameplay, and full state persistence using JPA/Hibernate with an H2 in-memory database. The project was developed using Java, Maven, JUnit, and object-oriented design principles.

Developed a collaborative computer vision project for drone control using hand gesture recognition. The system detects gestures such as fist, open hand, helicopter, and point-up gestures to control drone movement commands. The project focused on real-time image processing, gesture classification, and practical human–drone interaction. My contribution included working with the project codebase, testing gesture recognition behavior, and supporting the implementation/debugging process.