I am an early beta-user of the Synthyra PDE: an iterative, GA-driven pipeline for protein engineering that runs entirely on sequence-only models. It leverages diffusion pLMs to generate a massive starting population, then scores them through multiple cycles of selection using their custom pLM-based oracles. These oracles score for everything from PPI probability and binding affinity to standard stuff like foldability and solubility. For this specific run I set up the hyperparameters in an attempt to conduct many crossovers between the known Human protein interactors (Ephrin B2/B3) with the m102 antibody chains. I started with 10K variants went for 100 optimization steps. For the final screening I used Boltz-2 to calculate ipSAE in line with the competition's key metric.