Twelve domains, from transformer internals and token economics to agent architectures, LLMOps and AI governance. Written for architects, CIOs and engineers who need the how, not the hype. Every domain links to working demos in the AI Lab.
What the technology is, how it works under the hood, and what it really costs.
How models actually work: transformers, training pipelines, model families, context windows and how to read benchmarks without being fooled.
The economics layer: tokenization, pricing mechanics, latency anatomy, inference optimization and FinOps for the GenAI estate.
How production AI systems are actually designed and assembled.
Getting reliable behavior from probabilistic systems: system prompts, structured outputs, context budgets and memory architectures.
Grounding AI in your own knowledge: embeddings, chunking, hybrid search, GraphRAG and how to evaluate whether retrieval is telling the truth.
Autonomous systems done safely: tool calling, MCP, multi-agent patterns, planning, human-in-the-loop and guardrails.
When prompting isn't enough: the adaptation ladder, LoRA and PEFT, distillation, synthetic data and data readiness.
The new SDLC: coding agents, spec-driven development, vibe coding vs production engineering, and governing AI-generated code.
Platforms, integration, and keeping AI systems reliable in production.
The runtime estate: anatomy of an LLM API, model gateways, platform landscape, reference architectures and enterprise integration patterns.
Operating AI in production: evals as the new unit tests, tracing, versioning, hallucination analysis and CI/CD for AI systems.
Risk, regulation, strategy, and what's coming next.
Trust as an architectural property: the GenAI threat model, prompt injection, EU AI Act, NIST AI RMF, model risk and red-teaming.
From pilots to P&L: use-case prioritization, operating models and CoEs, AI-first enterprise architecture, value realization and adoption.
What's next and what it changes: reasoning and test-time compute, realtime AI, agent-to-agent protocols, world models and the quarterly watch list.