The LFC project - Learning From Commits - fits into the broad area of Just-in-time Software Defect Prediction (JIT-SDP), which seeks to use advanced algorithms to detect software changes (e.g., git commits) with the potential to introduce defects. The main challenge of JIT-SDP is dealing with data sent in streams, sparsely labeled with an unknown probability distribution and subject to concept drift. The LFC project, developed in partnership with UFRPE (Federal Rural University of Pernambuco) and the software company FAST, aims to create robust models to predict critical changes in software by developing a new predictive approach that is capable of learning from labeled and unlabeled change data. The solutions found by this project, when integrated into software version control systems, will provide early, reliable, and automated alerts of defect-inducing changes throughout the life of software projects. The LASSE team collaborates with the development of neural network model architecture, analysis of neural network optimization algorithms, and software change classification algorithms.