Accurate estimation of trip duration using New York City taxi data is crucial to optimize urban transport systems and improve taxi service operations. This study compares the effectiveness of four standard regression models—Decision Tree Regressor, LightGBM Regressor, Random Forest Regressor, and Gradient Boosting Regressor—and then explores feature engineering methods to create four composite models and discusses their predictive capabilities. The models are compared on the basis of Mean Squared Error (MSE), Mean Absolute Error (MAE) on the training and test datasets to analyse model performance. Our experiment results show that custom models consistently outperformed their baseline counterparts, with the Random Forest achieving the best performance and other custom models also demonstrating notable improvements. In order to make the experiments reproducible and comparable, we track and log all experimental (hyper)parameters, performance metrics and models using MLflow tool which manages the full lifecycle of ML projects.