This study presents recent advances in integrating geomechanics domain knowledge and neural networks, i.e. knowledge-driven neural networks (KDNN), for computational geomechanics. Two bespoke KDNNs for stress point integration and the discovery of the constitutive relations of geomaterials, will be introduced. The first part of this study revisits stress point integration from the perspective of neural networks, with the aim of simplifying conventional approaches [1]. This bespoke NN is proposed to compute the stress–strain response in the plastic regime—without requiring labelled data or the calculation of the Jacobian matrix—while still enforcing the consistency condition. The results demonstrate that this NN-based integration method accurately captures the elastoplastic stress–strain behaviour of both J2 and modified Cam-clay plasticity models, exhibiting strong robustness to variations in strain increment. Moreover, the approach is incorporated into a finite element method framework to simulate a three-dimensional bar under tensile loading, successfully reproducing the displacement, stress, and strain fields. The second part of the study focuses on the discovery of constitutive relations from data using a thermodynamics-informed learning framework, termed t-PiNet [2]. The fundamental relations of hyperelasticity and hyperplasticity theories are explicitly enforced through the network architecture, while physical constraints, including non-negative energy dissipation, are incorporated into the loss function. t-PiNet enables the automatic identification of internal variables and constitutive relations without prescribing a specific constitutive form a priori. Numerical examples under various cases demonstrate that the learned models exhibit strong generalisation capability and maintain physical interpretability. Overall, this study highlights the potential of KDNN to bridge conventional geomechanics knowledge and modern neural networks. The underlying concept of KDNN is general and holds promise for the development of a wide range of innovative methods and applications in geotechnics.