Building complex models from available data is a challenge in many domains, and in particular in food science. Numerical data are often not enough structured, or simply not enough to elucidate complex structures : human choices have thus a major impact at various levels (data and model structuration, choice of representative scales, parameter ranges, uncertainty assessment and management, expert knowledge). LIDeOGraM is an interactive modelling framework adapted to cases where numerical data and expert knowledge have to be combined for building an ecient model. Exploiting both stand-alone evolutionary search and visual interaction with the user, the proposed methodology aims at obtaining an accurate global model for the system, balancing expert knowledge with information automatically extracted from available data. The presented framework is tested on a real-world case study from food science : the production and stabilisation of lactic acid bacteria, which has several important practical applications, ranging from assessing the ecacy of new industrial methods, to proposing alternative sustainable systems of food production.