Resistance antibiotics among bacteria is on the rise. Research on finding therapeutics against antibiotic-resistant bacteria is currently active. One promising direction involves building over known antimicrobial peptides that constitute the host's innate immunity. There are now growing databases of antimicrobial peptides discovered on several species, but such peptides do not currently present suitable therapeutic template for several reasons, including toxicity at the concentrations needed to effectively kill bacteria. Given these constraints, it is of great interest to design novel antimicrobial peptides. Many laboratories that take on this task proceed via mutagenesis, introducing mutations in a systematic manner over a small family of known antimicrobial peptides. Due to its labor, the process is limited to small families and yields small libraries. In our laboratory, we have built a body of work that focuses on learning the relationship between sequence and antimicrobial activity. We have done so via machine learning models. Our models have gotten better over the years and have recently culminated in a deep learning model that currently outperforms many known models. Given a peptide sequence, the model assigns an antimicrobial activity score to a sequence that can be used to make a determination on the presence or absence of activity. In this work we show how such a model can be used in a generative setting to find novel antimicrobial sequences. Specifically, we couple the model with explorative algorithms that explore the sequence space of a template peptide. The model is used to score generated sequences and guide the exploration towards regions of the sequence space likely to contain active sequences. Visualization and analysis of the explored sequence-activity landscape show regions of antimicrobial activity that present rich data for further evaluation and refinement in the wet laboratory.