This study presents the fabrication and comprehensive characterization of ZnO/ZnS core–shell nanostructures grown on flexible polyimide (PI) substrates for dual-mode photodetection and gas sensing applications. The PI-based devices exhibited remarkable broadband photoresponse, achieving sensitivities of 100.52 (green), 113.24 (blue), and 1233.05 (UV) with corresponding responsivities of 3.92, 36.80, and 3287.18 μA/mW, respectively, and fast rise/recovery dynamics (e.g., 20.46/20.07 s in green and 16.31/24.11 s in UV), high gas sensitivity, and excellent mechanical flexibility. Sulfurization of ZnO nanorods into a conformal ZnS shell significantly enhanced visible-light responsivity, particularly under red and green illumination. The sensors demonstrated strong and selective responses toward CO, H₂, NO₂, NH₃, ethanol, and acetone under both dark and light-assisted conditions, where white-light photoactivation further improves the sensing response compared with dark operation (e.g., ZnS achieves NH₃ SI = 0.4067 and Resp = 11.69 μA/mW under white light) and is particularly attractive for energy-efficient, low-power activation using compact white LEDs, thereby providing a practical alternative to UV-assisted sensing on PI substrates under a constrained energy budget Mechanical bending tests further confirmed the robustness and operational stability of the devices under repeated deformation, where the unbent device delivered an average photocurrent of 3740.68 μA with a sensitivity of 37,406.76 and a response rate of 1453.68 μA/s, while maintaining high performance under mild bending (Bending 1–3). In addition, artificial intelligence (AI) techniques were integrated into the sensing system, where a one-dimensional convolutional neural network (1D-CNN) was employed to process dynamic sensing signals, enabling accurate gas classification and improving overall selectivity (e.g., 100% training accuracy for CO, NO₂, ethanol, and acetone under white light; and near-perfect UV performance with only one NO₂ misclassification; testing results showed ≥156 correct predictions per class under white light). These results underscore the potential of ZnO/ZnS heterostructures on PI substrates for next-generation wearable and energy-efficient sensing platforms with AI-enhanced functionality.