Fiber-reinforced biocomposites made from natural plant fibers and a biodegradable polymer matrix present a high-performance yet more sustainable alternative to conventional synthetic composites. Two microstructural aspects governing the composite behavior are the fiber orientation distribution and the fiber-matrix interface behavior, both of which are rather challenging to incorporate into a predictive modeling framework, particularly when aiming for computational efficiency. The stiffness homogenization is performed in the framework of continuum micromechanics (mean-field homogenization). At the composite scale, any fiber orientation distribution and any fiber aspect ratio distribution may be considered, such that biocomposite microstructures with more or less aligned fibers of different lengths can be suitably represented [3]. Fiber-matrix bonds are modeled by enriching the corresponding classical Eshelby problems by spring-type interfaces [1]. A model for the microstructure of the fiber itself is incorporated to homogenize the fiber stiffness based on the known stiffness of its intrinsic constituents (mostly cellulose and lignin) and on geometric features such as the microfibril angle and the lumen porosity [2]. The multiscale model is comprehensively validated by comparing the predicted composite stiffness to experimental results from uniaxial tensile tests, performed on several different plant fiber composites with various orientation distributions. To increase computational efficiency and make the model suitable for non-linear material models or even structural models, a neural network-based surrogate model is trained based on the model predictions. We chose a partially Bayesian Kolmogorov-Arnold network (KAN) architecture, a robust and high-performance alternative to classical multilayer perceptron (MLP) networks, that also yields uncertainty estimates for each output stiffness. [1] F. Dinzart, H. Sabar, and S. Berbenni. H. Int. J. Eng. Sci., 100:136–151, 2016. [2] M. Königsberger, M. Lukacevic, and J. Füssl. Struct., 56(1):13, 2023. [3] M. Königsberger, V. Senk, M. Lukacevic, M. Wimmer, and J. Füssl. Composites Part B: Engineering, 281:111571, 2024