Materials informatics utilizes data to establish relationships between the structures and properties of materials, enabling the exploration of the vastness of the materials space through the use of models. Trained on diverse datasets, the generative models can unlock the potential for predicting novel materials with tailored properties. However, while the key advantage of generative models is a potential to produce novel materials, often times they may be “too novel” and cannot be synthesized. In order to minimize the time and resources for experimental synthesis attempts, models that can predict the synthesizability (and, if synthesizable, synthesis recipes as well) would be immensely helpful. Thus, in this talk, I will delve into two important aspects of materials design: generation and synthesis prediction based on data and machine learning. I will also present the results of using large language models (LLMs) as strong baseline for synthesizability predictions and precursor selection problems. LLMs can also offer explanations for why certain materials are predicted as synthesizable while othere as unsynthesizable.