The proposed framework unites deep neural networks (DNNs) together with multi-objective optimization for designing environmentally friendly concrete mixes. A DNN model receives training through a wide dataset which includes multiple mix parameters along with curing conditions for accurate compressive strength prediction. The Bayesian hyperparameter tuning technique produces an optimal network configuration which delivers an average $$R^2$$ of 0.936 together with an RMSE of 5.71 MPa during 5-fold cross-validation. The Multi-Objective Particle Swarm Optimization (MOPSO) algorithm finds multiple optimal solutions which simultaneously optimize three competing objectives that include strength maximization and cost minimization and cement reduction. The optimized mix designs surpassed 50 MPa compressive strength through cement reduction of up to 25% which led to a total cost reduction of 15% compared to standard mix designs. The analysis of feature importance shows cement content together with concrete age serve as the main factors that affect strength measurements. The integrated data-driven method provides reliable decision-support tools to practitioners who need cost-effective sustainable mix designs through its identification of feasible trade-offs. The proposed methodology delivers new understandings of green concrete technology through optimal proportion discoveries that boost strength and save costs while decreasing environmental impact for direct application in real construction settings.