Patient-specific blood-flow simulations in cerebral aneurysms are computationally expensive. In practice, one is often interested in evaluating many scenarios, including different aneurysm geometries as well as variations in inflow boundary conditions such as heart rate and inflow magnitude. This results in a large number of simulations that are repeatedly initialized from poor initial guesses, making the transient start-up phase a significant computational bottleneck. We present a U-Net--based neural operator for blood-flow prediction in cerebral aneurysms, trained on data generated with Lattice Boltzmann Method (LBM) simulations using a shear-dependent non-Newtonian viscosity model. The training geometries are derived from an openly available cerebral aneurysm dataset, which we use to fit a MeshLab-based aneurysm generator capable of synthesizing anatomically realistic and diverse aneurysm shapes. Variability in inflow boundary conditions is incorporated by sampling global parameters such as heart rate, inflow magnitude, and velocity profile. Global inflow parameters are injected into the network via Feature-wise Linear Modulation (FiLM), enabling conditioning on patient- and scenario-specific boundary conditions. The model is trained on 2D slices of the 3D domain with time-varying periodic flow and predicts the full temporal evolution of the velocity field in a single forward pass, effectively learning a time-dependent flow operator. The objective is to significantly reduce the time required to initialize patient-specific blood-flow simulations. By providing an accurate initial flow field, the learned neural operator acts as a preconditioner for subsequent LBM computations, accelerating convergence from poor initial conditions and enabling fast, on-demand simulations for patient-specific aneurysm geometries. The proposed approach is part of a broader effort toward data-driven surrogate models for cerebrovascular applications, including transport processes and device-related simulations in complex aneurysm geometries.