In seismic hazard assessment, reliance on Vs30 proxies and 1D shear wave velocity profiles often leads to underestimated ground motion. This is particularly evident in areas with complex geological structures, such as Greater Beirut (GB). The metropolis is located near active seismic faults, with significant earthquakes in 551, 1202, and 1837. It is also characterized by a diverse range of soil compositions, from sandy terrains to limestone formations, and thus demands a detailed geotechnical modeling of the subsurface to constraint the seismic hazard. Our research has developed a comprehensive 3D geotechnical model for GB, integrating data from approximately 500 geotechnical boreholes, 700 geophysical measurements, a refined DEM and established geological insights. This model delineates variations in bedrock elevation and geological strata, with some sites exhibiting sediment depths up to 70 meters. Iterative data assimilation, combining H/V measurements with borehole data, has been fundamental in estimating the average shear wave velocity propagation (Vs-mean) throughout the model. To address data gaps in the southern part of GB, we employed a Random Forest machine learning model, trained on interpolated points from Kriging in the central part of the model. This approach ensured a continuous and comprehensive representation of subsurface conditions, even in areas with limited data. Building on this foundation, our ongoing work involves detailed seismic simulations to predict ground motion amplification in Beirut. Using a 3D hexahedral mesh generated via open-source Python code, we will conduct full 3D numerical simulations of seismic wave propagation using the spectral element method ( Komatitsch and Tromp, 2002; Komatitsch and Tromp, 2002; Komatitsch et al., 2023). These simulations aim to provide valuable insights into the seismic response of Beirut’s subsurface, contributing to the city’s earthquake preparedness and risk mitigation strategies. The conference presentation will showcase the first seismic scenarios based on our 3D geotechnical model, highlighting its potential in understanding and mitigating seismic risks.