Abstract: To achieve rapid detection of pesticide residues in tea, this study developed a dual-mode detection system that integrates bimetallic Au@Ag core-shell nanoparticles (Au@Ag NPs) with machine learning algorithms for surface-enhanced Raman spectroscopy (SERS) analysis. Three nanoparticle substrates were synthesized and evaluated, among which Au@Ag NPs were selected as the optimal SERS substrate based on their superior enhancement performance. Subsequently, SERS spectra of the target pesticides were acquired, and machine learning models were developed for both qualitative identification and quantitative determination. The proposed system achieved detection limits of 10⁻⁶ g/mL for carbendazim, 10⁻⁵ g/mL for clothianidin, and 10⁻⁶ g/mL for trichlorfon. For qualitative discrimination, the competitive adaptive reweighted sampling-random forest (CARS-RF) and improved variable selection optimization-random forest (IVSO-RF) models accurately classified the three pesticides, attaining test-set accuracies of 94.7% and 95.3%, respectively. For quantitative prediction, the CARS-RF-PLS model exhibited the best performance, achieving prediction-set correlation coefficients exceeding 0.9 for all pesticides and enabling accurate low-concentration quantification. This work effectively addresses the challenge of detecting trace-level multi-pesticide residues in the complex tea matrices and provides a reliable and rapid analytical method with reference value for related SERS-based detection applications.