Wind tunnel testing (WTT) remains a gold standard for aerodynamic validation, particularly at the boundaries of the flight envelope where Computational Fluid Dynamics (CFD) often shows limitations. However, these experiments are limited by the spatial coverage of pressure measurements. Historically, established methods like Gappy Proper Orthogonal Decomposition (Gappy POD) [1] have been widely used to fuse sparse measurements with high-dimensional simulation data. Recent advances in Deep Learning, specifically Graph Neural Networks [8, 9] and Implicit Neural Representations (INR) [3, 4], offer new opportunities to handle unstructured meshes and extract compact latent variables from physical fields. Notable applications include non-linear dimensionality reduction and aerodynamic pressure calibration [10]. The objective of this work is to extend sparse WTT data to the entire wing surface by leveraging the physics captured by CFD. This study introduces FusLaB (Fusion Latent Based), a novel methodology based on a conditional INR architecture. The framework addresses the limitations of linear models by summarizing geometry and flight conditions into a compact representation. The methodology relies on several key technical components. Firstly, a conditioning mechanism based on the Feature-wise Linear Modulation (FiLM) algorithm [5] modulates the decoder's layers via a hypernetwork. Global parameters and snapshot-specific latent vectors are jointly optimized using the CAVIA second-order meta-learning algorithm [6]. In order to capture sharp pressure gradients such as transonic shocks, the architecture incorporates multiscale Fourier Features [2] and an adaptive subsampling strategy during training. Validation was performed on the Airbus XRF1 transport aircraft configuration using data from the European Transonic Windtunnel (ETW) [7]. Results demonstrate that FusLaB significantly surpasses Gappy POD, particularly in capturing sharp pressure variations associated with shocks. The model maintains high fidelity even when the number of sensors is reduced to only 30 pressure taps. This approach provides a powerful tool for generating reliable aerodynamic datasets during early design phases. Future work will focus on multi-wing generalization to evaluate unseen geometries efficiently. References in attached document