In recent years, job recommender systems (JRSs) have been developed to compute matching scores of workers for projects. However, these JRSs do not satisfy monotonicity, which guarantees that if the skills of the worker improve, the matching score increases, and if the required skills of the project increase, the matching score decreases. Satisfying monotonicity is important because it is a matter of course for users of JRSs. Additionally, for JRS applications, it is desirable to compute the features of worker and project separately because we can add various ingenious devices to each feature extraction model. In the previous works, there are general purpose neural networks that guarantee monotonicity. However, they do not compute the features (or latent expressions) of two different types of input data separately. In this study, we propose a monotonic JRS that computes the features of worker and project separately. Our model consists of two processes: 1) computing the features of the worker and the project separately by the conventional monotonic neural networks; and 2) computing the matching score from sufficiency rates of the worker's features to the project's features. We can prove monotonicity of the entire model because the above processes satisfy monotonicity. From the experimental results, we show that our model has the minimum degree of accuracy. To improve this, we consider that it will be effective to replace the neural networks for computing the features with ones that have more expressive power.