The production of cement accounts for approximately 7–8% of global CO2 emissions, making it one of the largest industrial contributors to the greenhouse effect. Despite its significant carbon footprint, concrete remains indispensable due to its durability, versatility, and availability. Reducing its environmental impact requires innovative formulations with lower clinker content and/or alternative raw materials. Such substitutions, however, often alter mechanical performance, durability, and workability, complicating optimization and increasing R&D costs. In this context, artificial intelligence (AI) and data-driven modeling offer powerful tools to accelerate formulation development. By analyzing large datasets of existing concrete recipes, it becomes possible to identify trends and build predictive models for low-impact mixes. In this study, a dataset of 11 430 mixes from 1 272 recipes was analyzed, including binder composition, water-to-cement ratio, aggregates, and admixtures as inputs, and compressive strength and workability as outputs. After data cleaning, normalization, and feature selection, the dataset was split 80:20 into training and test sets. The model, optimized via Bayesian hyperparameter optimization using Optuna, achieved an R2 of 0.941 and MAE of 1.006 on the training data, and maintained good generalization (R2 = 0.670, MAE = 3.375) on the test set, demonstrating the potential of AI to support sustainable concrete design.