Multiscale modeling is an elegant and appealing approach to carry out in silico design, bridging gaps between time- and length scales for materials simulations from electrons and atoms to devices. Its practical realization was an area of intense research for several decades, with numerous success stories. At the same time, an understanding of significant challenges has developed.  In this talk, we focus on first-principles calculations of materials parameters, often used as the initial step in multiscale modeling. The calculations are often performed at temperature T=0 K, while obtained parameters are passed to higher-level models and used in simulation at realistic conditions at which materials operate in devices. Considering Ti, its alloys and compounds we argue that first-principles calculations of materials parameters for multiscale modeling should be performed by taking into consideration finite temperature effects, e.g. by means of ab inito molecular dynamics simulations (AIMD).  We demonstrate that AIMD greatly improves accuracy and reliability of a description of variety of materials properties and phenomena, ranging from mixing enthalpy to diffusion. Unfortunately, the AIMD simulations are quite time consuming. A highly promising scheme to improve the efficiency of AIMD simulations is provided by a combination of quantum mechanics calculations with machine learning techniques. Employing on-the-fly training of interatomic potentials described through moment tensors, we investigate temperature dependence of elastic moduli in several Ti-based systems and show that the calculated results are of the same accuracy as those obtained by AIMD with two orders of magnitude less computational effort. Collecting simulations results in databases of materials parameters provides additional opportunity for using reliable data calculated from first-principles in multiscale modeling.