-Context-aware application and services proposing potentially useful information to users are more and more widespread; however, their actual usefulness is often limited by the "syntactical" notion of context they adopt. The recently started AMBIT project aims to provide a general software architecture for developing semantic-based context-aware tools in a number of vertical case study applications. In this paper, we focus on the knowledge management foundations we are laying for the Semantic Engine of the AMBIT architecture. The proposed semantic analysis and similarity techniques: (a) exploit the textual information deeply characterizing both users and the information to be retrieved; (b) overcome the limits of syntactic methods by leveraging on the strengths of both classic information retrieval and knowledge-based analysis and classification, ultimately proposing information relevant to the user interests. The experimental evaluation of a preliminary implementation in an actual "cultural territorial enhancement" scenario already shows promising results.