Élivágár is a novel resource-efficient, noise-guided Quantum Circuit Search (QCS) framework designed to address the challenges of designing performant and noise-robust circuits for Quantum Machine Learning (QML). The framework innovates in three major aspects: search space, search algorithm, and candidate evaluation strategy. Élivágár generates circuits and data embeddings in a device topology-aware manner, avoiding the costly circuit-mapping co-search. It introduces two cheap-to-compute predictors—Clifford noise resilience and representational capacity—to decouple noise robustness and performance evaluation, enabling early rejection of low-fidelity circuits and reducing evaluation costs. Élivágár also searches for optimal data embeddings, significantly improving performance. Evaluations on 12 real quantum devices and 9 QML applications show that Élivágár achieves 5.3% higher accuracy and a 271× speedup compared to state-of-the-art QCS methods. The framework's key contributions include a novel QCS workflow tailored to QML systems, addressing design flaws in current classically-inspired QCS methods, and demonstrating significant improvements in circuit performance and resource efficiency.Élivágár is a novel resource-efficient, noise-guided Quantum Circuit Search (QCS) framework designed to address the challenges of designing performant and noise-robust circuits for Quantum Machine Learning (QML). The framework innovates in three major aspects: search space, search algorithm, and candidate evaluation strategy. Élivágár generates circuits and data embeddings in a device topology-aware manner, avoiding the costly circuit-mapping co-search. It introduces two cheap-to-compute predictors—Clifford noise resilience and representational capacity—to decouple noise robustness and performance evaluation, enabling early rejection of low-fidelity circuits and reducing evaluation costs. Élivágár also searches for optimal data embeddings, significantly improving performance. Evaluations on 12 real quantum devices and 9 QML applications show that Élivágár achieves 5.3% higher accuracy and a 271× speedup compared to state-of-the-art QCS methods. The framework's key contributions include a novel QCS workflow tailored to QML systems, addressing design flaws in current classically-inspired QCS methods, and demonstrating significant improvements in circuit performance and resource efficiency.