Unit 4 · AI and Domain Applications
The first three units developed skills in analysis, delivery, and code maintenance. This unit introduces two further constraints: model outputs are uncertain, and specialist settings impose additional requirements on data and evidence. General analysts can learn to check AI workflows; clinical and pharmaceutical readers can extend that work to statistical tables, traceability, and environment governance.
Begin with LLM conversations and structured output, explore tool calling, skills, and retrieval, then move to clinical reporting and regulated environments. Model-service examples need the appropriate accounts, permissions, credentials, and network access. Without those, study the interfaces and evaluation designs first; static review is not evidence of a successful service call.
Deliver a workflow with explicit inputs, processing steps, checks, and limitations. Distinguish program calculations, model-generated content, and human verification. Keep asking: Which conclusions have traceable sources? How will failures be detected? Who approves the results for use? These questions apply to ordinary reports as well as AI-assisted analysis. Course examples do not constitute organizational validation or regulatory approval.