The formulation of printable ceramic inks for additive manufacturing via direct ink writing remains a complex and time-consuming task, as it requires experimentally tuning the composition and rheological properties of the ink to ensure its printability . This process is typically based on trial and error, increasing costs and material waste.

In this work, we present the first stage of a data-driven formulation system built upon a hybrid information extraction pipeline that combines regular expressions with named entity recognition based on language models. The goal is to systematically retrieve key formulation parameters from full-text scientific articles. The pipeline identifies relevant entities such as powder composition, binder types and content, and water percentage, viscosity, yield stress, and viscoelastic moduli. A manually curated subset was used to validate the system, which achieves an 80% entity recognition rate. This strategy offers a promising tool to accelerate the design of new ceramic ink formulations for 3D printing, while significantly reducing manual effort, experimental costs, and material consumption. This work lays the foundation for a fully artificial intelligence AI-driven formulation assistant, where missing parameters can be inferred through predictive models and integrated into a structured database to support automated ink design.

This research work has been funded by the European Commission – NextGenerationEU, through the Momentum CSIC Programme: "Develop Your Digital Talent"