AI foundations and design
Understand learning, attention, multimodality, and model adaptation.
Choose model capabilities and explain their limits.
Chapters
Understand neural networks and language models through their objectives
Connect training objectives and token prediction to model behavior, use a softmax experiment to understand aggregation, and distinguish the full architecture from application evaluation.
Turn model capabilities into testable task contracts
Define capabilities, context, outputs, and abstention for classification and extraction, then verify facts, provenance, and execution permissions beyond valid structured output.
Handle images, speech, and video inside an application
Separate teaching-media processing into modality tasks, input checks, timing, assets, and review, with explicit evidence limits for sampling, generation, and recognition.
Decide between fine-tuning and application changes
Compare prompting, retrieval, fine-tuning, and serving through classification failures, including leakage, LoRA assumptions, memory estimates, and recovery of related model artifacts.
Define execution responsibility when connecting tools and agents
Distinguish MCP tools, A2A tasks, SDK loops, and skill resources, then use a narrow read contract to check authorization, parameters, and responsibility for results.
Build knowledge applications whose failures can be located
Separate knowledge applications into ingestion, retrieval, generation, and audit, checking provenance, permissions, abstention, and framework state to locate the failing stage.