We first used Boltzgen at a low batch size to determine the target hotspot and binder length with high success rate. Then we selected a suitable optimized approach and generated more binder backbones via RFdiffusion3 and Odesign to increase diversity. Following sequence design with ProteinMPNN/LigandMPNN and structure prediction via AlphaFold3/Protenix, we iterated the pipeline and scaled up the design batch size. Final filtering of high-quality predicted structures (with both high ipTM and ipSAE) was performed using Rosetta-ddg.