The eight-stage AI system lifecycle, the seven core AI engineering disciplines, a reference architecture for a generative-AI application, designing for AI-specific failure modes, and the versioning discipline unique to AI systems.
A production AI system is not a model with an API wrapped around it — it is an end-to-end architecture spanning problem definition, data collection and preparation, model development, evaluation, deployment, monitoring, and retrain and continuous improvement, the exact eight-stage lifecycle this studio's own free System Architecture map lays out. This module reframes everything the program has covered so far, from Module 1's fundamentals through Module 12's deployment practice, as components inside that larger architecture, and names the seven distinct engineering disciplines — data engineering, modeling and training, evaluation and governance, deployment and MLOps, observability and monitoring, infrastructure and platforms, and application and integration — a mature organization staffs to cover it.
By the end of this module you should be able to sketch a reference architecture for a generative-AI application, explain why AI systems fail differently than deterministic software and need explicit fallback design, and articulate why model version, prompt version, and data version each need independent tracking alongside code version — the versioning discipline Module 14's safety coverage assumes this module has already established.