For Generation IV nuclear reactors, optimal sensor positioning and real-time estimation of the quantities of interest are both open problems. In particular, the harsh environment of fast reactors, both due to the high radioactive levels and the presence of non-conventional coolants such as liquid metals or molten salts, is such that if possible, in-core sensor positioning requires careful attention. This problem is exacerbated for the Molten Salt Fast Reactor, which foresees fuel and coolant homogeneously mixed in the liquid phase and whose current design does not envision in-core solid structures. Thus, the possibility of estimating relevant in-core quantities, such as the neutron flux, from measurements taken outside the reactor core (for example, by sensors located in the reflector) is worth exploring, as it has important implications for safety, monitoring, and control. In this context, the Data-Driven Reduced Order Modeling framework offers a promising tool for combining out-core sparse measurements with some mathematical background knowledge, in the form of a reduced order model, on the in-core state to efficiently and accurately reconstruct the former in the whole core domain. This work explores this possibility by employing the Generalized Empirical Interpolation Method to retrieve the in-core neutron flux starting from sparse out-core noisy measurements, including a preliminary step of optimization of the sensor positioning in the reflector surrounding the core. The reconstruction capabilities, in both interpolation and extrapolation regimes, of the algorithm are really promising, showing that with few sensors, it is possible to infer significant information about the dominant physics inside the reactor core.