The plasma density is a crucial parameter of tokamak discharge as it impacts the fusion power, stability limits, influences detachment onset and impurities transport. The electron density profile shaping affects the deposition location of the external heating power, thereby playing a role, for instance, in NTM control. Therefore, the real-time estimation of the density profile is essential for high performance tokamak operation. The real-time density profile reconstruction can be provided by the RAPDENS code (Blanken et al. 2018), which collects density measurements, like interferometers and Thomson scattering (Pastore et al. 2023) and combines them with the solution obtained by its 1.5D particle transport model using the Extended Kalman Filter technique. RAPDENS can be employed in various applications including real-time dynamic state observation, offline reconstruction of the electron density profile with diagnostics not available in real-time, density controller design (owing to its provision of analytical Jacobians), and fast offline density profile simulation. This contribution presents the improvements made to its predictive model by introducing different boundary conditions, in particular a non-zero, time-varying density at the separatrix. Incorporating the dependence of separatrix density on gas fueling in the non-zero boundary condition improves the quality of the offline simulations used for controller design. Furthermore, this adjustment enhances the versatility of the RAPDENS predictive model, allowing for more efficient density profile reconstruction across a broader range of discharge scenarios and reducing the effort needed to find a suitable combination of tuning parameters to match the density profile at the edge. A method using a real-time compatible empirical formula (Kallenbach et al. 2018) to determine the separatrix density as a boundary condition for AUG discharges is proposed, and the results of offline RAPDENS simulations are compared with the density profile reconstructed using integrated data analysis (IDA) (Fischer et al. 2010).