The pipeline consists of a feedback look controlled by an agent that at each iteration makes decision for the next iteration based on the results of the previous one. One iteration consists of the following: A structure generation using RFdiffusion providing hotspots and CDRs lengths. An iteration of proteinMPNN and solubleMPNN for different regions of the created binder followed by validation with ablang2 perplexity and scoring where ipsae, podckq, dSASA, deltaG per res, unsaturated H bonds and sc are scored. During scoring, the 3 best structures go through another round of proteinMPNN inverse folding that modifies specific residue positions based on their per residues results in the previously cited metrics. At the end of the generation, a report is created that is then fed back to the agent (llama 3 improved for bioengineering) to reason over hyperparameters for next round.