Micro-models refer to small, specialized machine learning models each designed to tackle a narrow subproblem within a larger system. Instead of having one large model handling everything, a system may be broken into components where each micro-model is responsible for a specific aspect of the data or a specific prediction task. For example, in an NLP pipeline, one might have separate micro-models for entity recognition, sentiment analysis, and so on, whose outputs are combined. In the context of a larger AI system, micro-models can be thought of as analogous to microservices in software architecture: each is simpler, focused, and can be developed or updated independently. The benefits include modularity, easier debugging, and potentially more efficient training on small datasets for each task. In computer vision, one might have micro-models detecting different kinds of features or objects which then feed into a higher-level decision model. This concept appears in discussions of composability in AI, and in productization where multiple narrow models can be orchestrated to solve a complex problem.