Growth differentiation factor-15 (GDF15), a stress-responsive cytokine of the transforming growth factor-β superfamily, is elevated in cancer cachexia, chemotherapy-induced nausea, and hyperemesis gravidarum, making it both a biomarker and a therapeutic target. Here, we developed high-affinity GDF15 binders using an artificial intelligence-driven protein design framework. To achieve this, we systematically explored three complementary scaffold-generation strategies: scaffold grafting, diffusion-based de novo design, and scaffold-search and grafting, identifying distinct advantages - scaffold grafting rapidly optimized receptor-derived motifs to sub-nanomolar affinity; de novo diffusion produced topologically novel binders; and scaffold-search and grafting enabled access to concave site B of GDF15 by repurposing evolutionary structural analogs from natural complexes. The designed GDF15 binders were translated into two functional modalities. First, a one-step, wash-free luminescent biosensor was created by coupling a de novo binder to split-luciferase fragments, enabling the rapid and sensitive quantification of GDF15. Second, the highest-affinity binder was engineered as an Fc-fusion decoy receptor, thereby effectively neutralizing GDF15 signaling in cell-based assays (IC₅₀ = 7.2 nM), demonstrating comparable in vitro potency to ponsegromab, a monoclonal antibody currently undergoing phase II clinical trials. Together, this work establishes a versatile artificial intelligence-driven binder design pipeline with broad potential for next-generation diagnostics and therapeutics in cancer cachexia and other GDF15-mediated diseases.