A systems-first course teaching technical professionals how modern AI stacks work including tokenization, RAG, agents, cost estimation, model selection, deployment, and security
Tokens, RAG, agents, MCP.If you work anywhere near technology today, you’re hearing these terms in vendor pitches, architecture reviews, and planning meetings. Most people nod along — but nodding isn’t the same as understanding what’s actually happening under the hood.This course closes that gap. To set expectations: this is not a “ChatGPT tips” class. Instead, you’ll get a systems-first walkthrough that explains how the entire stack behaves — from tokenization and cost, through retrieval and agent loops, to deployment and operational risks.
Why this course mattersThese buzzwords don’t go away just because you don’t fully understand them. When you know the mechanics, you can answer the questions that actually matter in production: How much will this cost? Where will it fail? What attack surfaces exist? What trade-offs should we accept between latency, accuracy, and expense?
What you’ll get out of the courseKnowing how to prompt a chatbot is useful, but doesn’t prepare you for the operational decisions you’ll be asked to make. This course focuses on the practical skills that let you evaluate, compare, and secure real-world AI systems.By the end of this course you’ll be able to:
Explain what a large language model (LLM) is doing when it answers a prompt.
Estimate a real workload in tokens and convert that to an approximate dollar cost.
Select an appropriate model based on latency, throughput, accuracy, and cost.
Strengthen weak prompts into robust, production-ready prompts.
Ground a model in your own data using retrieval-augmented generation (RAG).
Trace how an agent executes its loop, including tool calls and MCP (Model Control Plane) interactions.
Identify common failure modes — hallucinations, prompt injection, or agents with excessive privileges — and decide when AI is not the right tool.
Summary of outcomes and examples
Skill
What you’ll do
Practical example
Token costing
Estimate tokens and cost
Calculate tokens per request × requests per month → cost
Model selection
Balance cost, latency, and accuracy
Choose a smaller model for low-latency lookup tasks
RAG implementation
Attach domain data to model answers
Build a search + re-ranker + generator pipeline
Agent tracing
Understand tool calls and MCP flows
Map the call sequence when an agent invokes a database tool
Security posture
Spot and mitigate failure modes
Implement prompt sanitation and least-privilege tool access
About the instructorI’m Bryan Krausen. I’ve taught technical professionals how the systems they depend on actually work — covering infrastructure as code, security, and automation. AI is now one of those foundational systems, and my teaching approach is the same: open the hood, trace the components, and give you the tools to reason about trade-offs and failure modes.
How we’ll learn
Slides (roughly half): precise concepts, definitions, and vocabulary so the ideas become repeatable and usable.
Lightboard sessions (roughly half): step-by-step visual breakdowns of complex ideas so the mechanics actually click.
Hands-on labs: short exercises where you’ll tokenize inputs, cost workloads, connect a live tool to a model, and inspect agent traces. This is where the course turns from “things you’ve heard” into “things you can do.”
If you haven’t seen a lightboard, it’s a whiteboard-style recording technique that lets me build diagrams and walk through them visually while we talk. The goal is to demystify the “black box” impression: we’ll open that box and examine each piece.There’s no code to install to get started. Whenever a term trips you up, use the downloadable glossary PDF that collects every key acronym and definition on a single page.
A concise glossary PDF is available to help when a term or acronym gets in the way of progress.
