AI Engineering
Build the deterministic engineering around language models — tokenizers, retrieval, MCP tool routers, and agents — running for real in your browser
A hands-on AI engineering course for the era after 'how do I call the API' became free. Instead of watching a model generate text, you build the infrastructure that makes AI features shippable: a byte-pair tokenizer, a retrieval-and-reranking pipeline, a Model Context Protocol (MCP) message router on the current 2026 spec, tool dispatch with sandboxing, multi-agent orchestration with failure recovery, token-cost engineering, prompt-injection defences, and an evaluation harness. 10 modules, 53 lessons, roughly 16 hours, with 378 exercises and quiz questions — every one deterministic JavaScript that executes and grades in your browser. No API key required.
- ✓ Tokens & Cost
- ✓ Retrieval & RAG
- ✓ MCP & Tools
- ✓ Agents & Safety
This course includes:
- 📹 16 hours of video content
- 📝 10 comprehensive modules
- 💻 Hands-on coding projects
- 🏆 Certificate of completion
- 🔄 Full access during your selected plan
Course Overview
What You'll Learn
Master these essential skills that employers are actively seeking
Tokens & Cost
BPE tokenization, context-window accounting, and a cost model you build yourself
Retrieval & RAG
Embeddings, chunking, reranking, grounding — and the failure modes of each
MCP & Tools
The Model Context Protocol implemented from JSON-RPC up, plus tool schemas, discovery, and sandboxing
Agents & Safety
Agent loops, multi-agent orchestration, excessive-agency limits, injection defences, and evals
Learning Outcomes
By the end of this course, you will be able to:
📋 Prerequisites
- Basic JavaScript (variables, functions, arrays)
- Familiarity with JSON
- No ML background and no API keys required — nothing in the course calls a live model
How It Works
Build the Primitives
Tokenizer, similarity search, chunking — the parts every AI system is made of
Wire the Protocols
JSON-RPC, the MCP handshake, tool schemas and dispatch — implemented, not described
Orchestrate Agents
State machines, supervisor-worker patterns, handoffs, and failure recovery
Ship Responsibly
Token economics, injection defences, guardrails, and a golden-trace eval harness
Tips for Success
Run every code block — the course is built around execution, and the surprises are the lesson
Module 2 (tokens and cost) pays for itself first: cost reasoning is the most common gap in AI teams
Treat the MCP router you build as a portfolio piece — extend it after the course
Module 9's savings scorecard is the argument that keeps AI features alive in production — internalise it
Frequently Asked Questions
No. The course deliberately never calls a live model: every exercise is deterministic JavaScript that runs in your browser. What you build — routers, retrievers, budgeters, evals — is the part of AI systems that is ordinary, testable software.
Because the durable, hireable skill is the deterministic infrastructure around the model. Calling an API is one line; the systems around that line are the job.
The Model Context Protocol module was written against the 2026 specification — two transports (stdio and Streamable HTTP), the initialize handshake, and capability negotiation — with the older retired transport taught as history.
Basic JavaScript. If you can write a function and a loop, you can take this course; our JavaScript course covers everything assumed.
Related Courses
Ready to Master AI?
Learn by building and running real code, one lesson at a time