Chia extract has been increasingly used in the food industry since it is rich in bioactive compounds, such as fatty acids, omega-3 fatty, antioxidants, proteins, vitamins, minerals, and dietary fiber. This extract can be obtained by using conventional extraction techniques (e.g., pressure) on chia seeds. Unfortunately, such techniques are insufficient to access all chemical components present in the seeds matrix, producing a by-product named chia cake that is usually discarded. On the other hand, since chia cake contains significant nutraceutical properties, it is still viable and beneficial to perform extractions of chia extract from chia cake. A typical objective of an extraction is to gather a high mass yield of chia (cake) extract. Since the extraction process is complex and expensive (e.g., in terms of laboratory resources), there is an increasing interest in determining the mass yield based on variables of the extraction like temperature, extraction time, and solvent. In this paper, we study the viability of applying traditional fuzzy inference systems (e.g., based on Mamdani's method) and adaptive neurofuzzy inference systems (ANFIS) for this problem. We propose a fuzzy inference architecture that predicts the mass yield of chia cake extract based on temperature, extraction time, and solvent. Our architecture makes use of fuzzy sets and fuzzy rules in the context of fuzzy inference methods. To design them, we create and use a dataset that contains the mass yield of real extractions conducted in the laboratory under different configurations. Hence, it represents another contribution of this paper and serves as the needed foundation to build the proposed architecture. Further, we conduct a performance evaluation to choose the fuzzy inference system that better fits the architecture. Based on our analysis, ANFIS was the best inference method since it delivered the lesser errors and greater correlations between predicted and observed values. We conclude that fuzzy inference systems are powerful tools for the food industry since they can capture the intrinsic imprecise nature of the extraction process, model the existing non-linear relations of the variables, and represent the expert domain knowledge.