Controlling the expression of genes is one of the key challenges of synthetic biology. Until recently fine-tuned control has been out of reach, particularly in eukaryotes owing to their complexity of gene regulation. With advances in machine learning (ML) and in particular with increasing dataset sizes, models predicting gene expression levels from regulatory sequences can now be successfully constructed. Such models form the cornerstone of algorithms that allow users to design regulatory regions to achieve a specific gene expression level. In this review we discuss strategies for data collection, data encoding, ML practices, design algorithm choices, and finally model interpretation. Ultimately, these developments will provide synthetic biologists with highly specific genetic building blocks to rationally engineer complex pathways and circuits.