In the context of smart farming, development of advanced analytics translating raw data into useful information for monitoring and decision support is crucial. Oftentimes, these analytics serve to estimate underlying states (e.g. health or welfare status) of a target subject (e.g. a farm animal) based on sensor time-series data. In an agriculture, these target processes are generally complex and influenced by a multitude of internal (e.g. genetics, physiology, history) and external (e.g. contexts, management) factors, rendering simple time series models insufficiently accurate. One solution is via the integration of expert knowledge in the analytics to achieve better interpretation of time-series data. In this project, I will build a data-analytics framework implemented in an open software library that explicitly integrates different sources of expert knowledge in a state estimation model. To this end, I will model (1) theory and domain expertise, (2) context and control measures, and (3) historical data patterns and incorporate these in a hidden Markov and other state-estimation models. Initially, the framework is developed with a focus on animal production based on 2 concrete case studies, but its implementation will allow for extension towards a variety of agricultural applications. As such, this project can play an important role in solving a timely and relevant problem in food production, and it contributes to a more sustainable and socially accepted agricultural sector.