Ehsan tabatabaei

Designed and deployed an end-to-end recommender system for a social media platform, handling multimedia content (video, audio, images). Built collaborative-filtering and content-based pipelines using Scikit-Learn and XGBoost; integrated an online learning loop to refresh model weights from new interaction data, improving click-through rate by ~15%. Owned the full MLOps lifecycle: data ingestion, feature engineering, model training, A/B testing, and production deployment.

Investigated deep learning approaches for channel estimation in underwater acoustic OFDM systems; designed DNN-based estimators that significantly improved accuracy in high-noise environments. Authored experimental code in Python/TensorFlow; benchmarked results against classical MMSE and LS estimators.

Computer Vision & Image Processing Intern Developed feature extraction algorithms for number recognition in noisy environments, achieving over 98% accuracy using an OpenMV camera. Focused on shape representation, description, and feature visualization techniques for robust industrial applications.