For mix design of Alkali-Activated Concrete (AAC) accurate compressive strength predictions are required with limited availability of experimental data. Traditional machine learning models base their reliability on the amount of data used to build them. Therefore, the generalization across diverse data of AAC mixtures utilizing different precursors and activators is an issue for these models. This study introduces a meta-learning approach, leveraging Model-Agnostic Meta-Learning (MAML) and Reptile, to enable the rapid development of a predictive model for new AAC mix design. Using initially only a limited amount of data and training the model on a variety of tasks defined by AAC mix properties and curing conditions, MAML learns an optimal initialization, facilitating few-shot learning. This allows efficient fine-tuning with available data, offering enhanced adaptability and generalization across new AAC formulations. Our results demonstrate the model’s potential for real-time compressive strength predictions, optimizing both research and industry applications by reducing experimental workload and costs.