Abstract Predicting the nonlinear dynamics of synaptic transmission is a fundamental challenge in computational neuroscience and neurodynamics. Although the prediction of biological signals, notably field excitatory postsynaptic potentials (fEPSP) in the hippocampus, has conventionally relied on generic, slice-agnostic architectures, reservoir computing (RC) has emerged as a powerful paradigm for modeling neural activity and offers considerable potential for achieving high-fidelity predictions in this task. In this paper we analyze the peculiarities of optimizing RC for predicting fEPSP responses in the CA3 and CA1 regions of mouse hippocampal slices elicited by stimulation of the dentate gyrus (DG). We demonstrate that substantial improvements in predictive performance can be realized by tailoring the reservoir architecture and its hyperparameters, including spectral radius, input scaling, and reservoir size, to the unique dynamical signatures of individual hippocampal slices. Our results reveal that optimized reservoirs significantly outperform non-optimized counterparts by effectively capturing slice-specific heterogeneities inherent in the DG–CA3–CA1 circuit dynamics. These findings underscore the critical importance of personalized architectural design in RC for biological time series prediction. Moreover, they point toward a promising pathway for developing more accurate models of hippocampal synaptic function, constructing virtual detailed hippocampal systems, and advancing neuromorphic neuroprosthetic technologies.