New forms of computation are increasingly emerging directly within physical devices, enabling embedded processing of external stimuli and sensor data. Physical reservoir computing is a hybrid computational paradigm in which a physical system (the reservoir) exhibits nonlinear dynamics capable of capturing temporal correlations between present and past inputs, similarly to an analog recurrent neural network [1, 2]. The reservoir nonlinearly projects input signals into a high-dimensional feature space, where a trainable linear readout layer is used for task-specific predictions. We propose an in-materia computing platform based on a simple doubly clamped thin aluminum beam acting as a geometrically nonlinear oscillator. Under appropriate modelling assumptions, the system dynamics are described by a Duffing oscillator that captures the nonlinear coupling between axial and bending behaviors. Although the setup consists of a single-degree-of-freedom dynamical system, the effective number of computational nodes is increased through a time-multiplexing strategy, without modifying the intrinsic beam dynamics or introducing delayed feedback mechanisms. The computing capabilities of the device are first investigated through numerical simulations to assess the interplay between the physical reservoir, the readout layer, and the target tasks. Several benchmark problems are considered, including parity function evaluation, signal classification, and image recognition, with reservoir dynamics simulated using Runge–Kutta integration. The proposed computational framework is finally validated experimentally with an aluminum 3D-printed beam, using an electrodynamic shaker for actuation and a laser Doppler vibrometer to measure its nonlinear response. This work demonstrates the feasibility of effective physical reservoir computing with minimal mechanical complexity, highlighting a scalable route toward mechano-intelligence.