The accurate delineation of hydrocarbon-bearing intervals is essential for efficient reservoir development, influencing decisions ranging from well placement to production optimization and resource evaluation. Traditional workflows employ seismic inversion to derive elastic-property volumes such as acoustic impedance (Ip) and the Vp/Vs ratio, followed by rock-physics-template interpretation to distinguish fluid effects (Ødegaard and Avseth, 2004). Although these methods are widely adopted, they introduce uncertainties at multiple stages. Inversion quality depends on accurate wavelet estimation, both in frequency content and phase alignment (Latimer et al., 2000), constructing the low-frequency background model often relies on sparsely spaced well logs (Sams and Carter, 2017) and assumptions of weak lithology contrasts (Avseth et al., 2016) or isotropy can neglect anisotropy and rock heterogeneity (Asaka, 2018). These factors undermine confidence in fluid predictions and risk costly misinterpretations. Artificial Intelligence (AI), and specifically Deep Learning (DL) circumvents many inversion pitfalls by learning direct mappings from seismic and geological inputs to subsurface properties. Convolutional neural networks (CNNs) can learn complex, non-linear mappings between seismic amplitudes and target attributes, capturing geological and geophysical context if appropriately trained (Zhang et al., 2021; Li et al., 2022). Such approaches can bypass explicit inversion approaches by directly inferring elastic or lithologic parameters from seismic and ancillary data. Synthetic studies suggest that deep learning can predict porosity, permeability, and elastic moduli with high accuracy under idealized conditions. Yet practical deployment for field-scale fluid detection has been limited by challenges in generalization, scarcity of representative training data, and difficulty in integrating with rock-physics theory. Heir and Aghayev (2025) started bridging the gap by applying a 1D Temporal Convolutional Network (TCN) to predict Vp, Vs, and Ip at well locations, and subsequently applying rock physics, demonstrating the important contribution of seismic data for DL prediction of blind-well estimates of fluid presence. However, their approach remained limited to individual well traces.