A semantic embedding is a logical encoding of a formal language, namely the object language, into the specification language of a logical framework. In their seminal paper “Experience with embedding hardware description languages in HOL,” Boulton et al. coined the terms deep and shallow embeddings depending on whether or not the syntax of terms of the target language is represented by a data type in the specification language. Thus, a deep embedding enables reasoning about classes of terms, while a shallow embedding limits reasoning to concrete terms. Embeddings of programming languages are well-known applications of interactive theorem provers, specially of those based on higher-order logic. These embeddings are often intended to support the study of a programming language semantics or to enhance a programming language with the deductive capabilities of the logical framework. A different type of embeddings, here referred to as structural embeddings, are intended to augment specification languages with structural elements of the object language. In a structural embedding, the outermost elements of the object language, i.e., the structural parts, are encoded, either deeply or shallowly, but the internal elements, i.e., the basic expressions, are those of the specification language. Advances in automated reasoning and user interfaces have enabled structural embeddings to enhance usability of interactive theorem provers and to reduce the gap between verification tools and modeling tools used by practitioners. This talk presents an overview of several years of research on theorem proving in safety-critical aerospace systems through the lens of embeddings and, more particularly, structural embeddings. It discusses lessons learned and provides examples of successful applications to automated reasoning, termination analysis, floating-point analysis, and, more generally, verification of cyber-physical systems. Our main point, which is hardly original, is that interactive theorem provers could serve as intermediate systems that connect a cluster of components. Structural embeddings could then provide the frontend capabilities to access this cluster of components.