Atomistic computer simulations are widely used to predict thermophysical properties. Recent advances in artificial intelligence have enabled the development of atomistic simulation models trained on large databases of ab initio (first-principles) quantum-mechanical calculations. However, simulations based on these models still struggle to reproduce experimental measurements accurately because they inherit the limitations of the underlying theoretical methods. At the same time, they largely overlook the vast amount of experimental data accumulated over decades. My team has recently introduced a novel methodology for developing hybrid models that combine ab initio and experimental data, and we have successfully applied it to specific substances, achieving very high accuracy. This approach opens new possibilities for constructing atomistic simulation models by enabling the systematic exploitation of the wealth of existing experimental data. In this talk, I will discuss the theoretical foundations of this methodology and present applications to two systems: water and copper. I will show how combining ab initio and experimental data leads to models that accurately capture the density and structure of these substances while also generalizing to other properties, such as melting temperatures. Finally, I will discuss the challenges and opportunities involved in scaling this approach towards a general framework for the accurate prediction of thermophysical properties across a broad range of substances.