AI Engineer in 12 Weeks

PRO

Build production-ready AI systems in 12 weeks using agents, RAG, evals, multimodal tools, and deployment workflows. Ship 3-4 portfolio-ready projects and one capstone with architecture decisions, eval results, and deployment evidence.

60 chapters240 lessons120 hours

Chapters

What AI-First Engineering Actually Means

Understand the fundamental shift from manual-first to AI-assisted engineering workflows. Learn to collaborate with AI tools while maintaining engineering ownership.

4 lessons40 min

Coding with Agents: Claude Code, Cursor, and Codex

Compare the major coding agent tools, understand their workflow styles, and learn to write effective prompts for each.

4 lessons50 min

Spec-Driven Development for AI-Assisted Building

Learn to write specifications that produce reliable, aligned AI outputs. Turn vague feature requests into structured engineering specs with clear acceptance criteria.

4 lessons65 min

Introduction to MCP

Understand the Model Context Protocol -- the open standard for connecting AI applications to external tools, data sources, and services.

4 lessons55 min

Week 1 Lab: Build an AI-Assisted Feature from a Spec

Pull together everything from Week 1 into a hands-on project. Write a spec, plan tasks, prompt an agent, build a feature, and reflect on the process.

4 lessons90 min

What an LLM App Actually Is

Demystify the architecture of LLM-powered applications. Understand the request lifecycle, the role of the API layer, and how prompts, context, and responses flow through a production system.

4 lessons50 min

Prompt Design for Reliable Engineering Outcomes

Move beyond casual prompting to systematic prompt engineering. Learn the techniques that make LLM outputs predictable, testable, and production-worthy.

4 lessons50 min

Structured Outputs and JSON Contracts

Learn to constrain LLM outputs to exact schemas using Anthropic's structured output features. Build reliable data pipelines where the model's output is guaranteed to match your application's expectations.

4 lessons50 min

Function Calling and Tool Use

Give your LLM applications the ability to take actions in the world. Learn Anthropic's tool use API, design safe tool interfaces, and build agents that call functions reliably.

4 lessons55 min

Week 2 Lab: Build a Ticket Triage Service

Apply everything from Week 2 in a capstone project. Build a production-grade ticket triage service that classifies, routes, and enriches support tickets using structured outputs and tool calling.

4 lessons60 min

What RAG Actually Is

Understand Retrieval-Augmented Generation -- what it solves, how it works, and when to use it instead of fine-tuning or long context windows.

Embeddings from First Principles

Learn how text becomes vectors, why semantic search outperforms keyword search, and how to choose the right embedding model for your RAG system.

Chunking and Indexing

Master the art and science of splitting documents into effective chunks -- the single biggest quality lever in any RAG system.

Building a Basic Retrieval Pipeline

Connect all the pieces -- indexing, querying, and grounded prompt assembly -- into a working end-to-end retrieval pipeline.

Week 3 Lab: Build a Grounded Q&A Assistant

Apply everything from Week 3 to build a complete RAG-powered Q&A assistant with corpus selection, chunking, retrieval, and grounded generation.

Why Basic Retrieval Fails

Understand the five failure modes that cause RAG systems to return wrong answers even when the right document exists in your corpus.

Hybrid Search: Dense + Keyword Retrieval

Combine dense vector search with keyword matching to handle queries where exact tokens matter as much as semantic meaning.

Reranking and Two-Stage Retrieval

Add a reranking stage to your retrieval pipeline that dramatically improves result quality by re-scoring candidates with a more powerful model.

Retrieval Evaluation and Quality Measurement

Build a systematic evaluation framework that measures retrieval quality independently from generation quality, enabling targeted improvements.

Week 4 Lab: Upgrade a Weak RAG System

Apply everything from Week 4 to diagnose, improve, and measure a RAG system's retrieval quality using hybrid search, reranking, and systematic evaluation.

Workflows vs Agents

Understand the fundamental architecture decision between orchestrated workflows and autonomous agents, and learn when each approach is the right choice.

Single-Agent Design with Tools

Learn to design effective single-agent systems with high-quality tools, clear stop conditions, and proper guardrails.

State, Memory, and Control Flow

Master the design of agent state models, control flow patterns, and memory systems that keep agents grounded and efficient.

When Not to Use Multi-Agent Systems

Learn to resist the multi-agent temptation and apply the single-agent-first rule to build simpler, more reliable systems.

Week 5 Lab: Build a Research-and-Action Assistant

Apply everything from Week 5 by designing and building a single-agent assistant with search, action, and stop conditions.

What Evals Actually Are

Understand why AI systems fail silently, what evaluations measure, and how to think about reliability as an engineering discipline rather than a hope.

4 lessons45 min

Building a Useful Eval Dataset

Learn to design, build, and grow evaluation datasets that actually catch the failures your system will encounter in production.

4 lessons45 min

Graders, Rubrics, and Scoring Strategies

Master the three grading approaches -- code-based, LLM-as-judge, and human review -- and learn to write rubrics that produce consistent, actionable scores.

4 lessons45 min

Regression Testing and Reliability Loops

Build the flywheel that catches regressions before they ship: baseline measurements, offline vs online evaluation, and the discipline of continuous comparison.

4 lessons45 min

Week 6 Lab: Build an Eval Harness

Combine everything from this week into a complete eval harness: dataset, graders, baseline, regression check, and production monitoring plan.

4 lessons50 min

How to Approach an Unfamiliar Legacy Codebase

Systematic strategies for understanding existing systems before adding AI capabilities -- mapping flows, finding safe entry points, and avoiding modernization pitfalls.

Integration Points, APIs, and Context Boundaries

Design precise interfaces between AI features and legacy systems -- controlling what the model can read, what it can do, and what stays off-limits.

Authorization, Permissions, and Data Safety

Implement least-privilege access, OAuth 2.1 patterns, audit trails, and compliance-aware permission models for AI systems in regulated environments.

Human-in-the-Loop and Safe Rollout

Design approval flows that pause AI execution for human validation, and plan staged rollouts that build trust incrementally from shadow mode to limited autonomy.

Week 7 Lab: Add an AI Feature to a Legacy System

Combine all Week 7 skills into a comprehensive project -- map a legacy system, design integration boundaries, implement permissions, add HITL gates, and plan a staged rollout.

What Multimodal AI Actually Changes

Understand when working across text, images, and audio genuinely improves a workflow -- and when it is a distraction. Learn to choose modalities based on the user's job, not API availability.

4 lessons55 min

Vision and Image Understanding Workflows

Build production-grade vision pipelines that extract structured information from images, handle failure modes gracefully, and include verification steps that make the output trustworthy.

4 lessons55 min

Speech, Audio, and Voice Interfaces

Build speech-to-text and text-to-speech pipelines that handle real-world audio conditions, design voice interfaces with proper correction flows, and evaluate when voice genuinely improves a workflow.

4 lessons55 min

Designing Good Multimodal Products

Learn to design multimodal products that start from the user's job, include correction paths for every modality, and measure workflow success instead of just API output quality.

4 lessons55 min

Week 8 Lab: Build a Multimodal Assistant

Design, build, and evaluate a multimodal AI assistant that uses vision, audio, or both to solve a real user problem. Produce a portfolio-ready project with spec, implementation, failure analysis, and reflection.

4 lessons60 min

From Prototype to Production

Understand what changes when real users arrive and build a production readiness discipline that separates demos from deployable systems.

Tracing, Telemetry, and Observability

Learn to instrument AI systems with distributed tracing and telemetry so you can see inside every request and diagnose problems in minutes, not days.

Operational Metrics and Failure Detection

Learn to choose metrics that catch real problems, design alert thresholds that avoid fatigue, and build a taxonomy of AI system failures.

Latency, Token Usage, and Cost Control

Master the economics of AI systems -- understand cost drivers, apply optimization levers, design latency budgets, and implement cost controls that prevent runaway spending.

Week 9 Lab: Productionize an Earlier Project

Apply everything from this week to productionize one of your earlier projects with a complete deployment plan, observability design, operational metrics, cost controls, and incident playbook.

When Fine-Tuning Is Actually the Right Move

Learn the optimization ladder and when model adaptation is -- and isn't -- the right engineering decision.

Supervised Fine-Tuning from a Product Perspective

Understand SFT as behavior shaping -- designing training datasets, evaluating results, and knowing when SFT delivers real value.

PEFT, LoRA, and Efficient Adaptation

Understand parameter-efficient fine-tuning methods that make model adaptation faster, cheaper, and more flexible.

Preference Optimization and Reinforcement Fine-Tuning

Explore advanced adaptation methods -- DPO and RFT -- that go beyond demonstrations to learn from preferences and rewards.

Week 10 Lab: Create a Real Adaptation Plan

Apply everything from Week 10 to create a comprehensive adaptation plan for a real system -- making sound engineering decisions about when and how to adapt.

Choosing the Right Studio Project

Select a project that demonstrates engineering judgment, scope it to its strongest minimal form, and plan milestones that build from risk to confidence.

Architecture Decisions Under Constraint

Learn to make and document architecture decisions that prioritize fitness over complexity, and communicate tradeoffs with clarity and confidence.

Building the Core System

Execute the core AI workflow first, add quality evidence through evaluation, and layer in safety and observability to demonstrate production readiness.

Polish, Demo, and Documentation

Transform your working system into a portfolio piece through clear documentation, structured eval evidence, and a demo that tells the story of your engineering judgment.

Week 11 Lab: Build Your Studio Project

Execute your studio project sprint from launch to submission, building the core system, adding quality evidence, and packaging everything for professional presentation.

What Makes a Strong Capstone

Transform your project from a demo into a professional engineering case study that proves judgment, not just skill.

Architecture Decision Records and Technical Communication

Make your engineering thinking visible through ADRs and precise technical writing that convinces skeptical peers.

Deployment Checklist and Release Readiness

Prove you think like an operator with deployment checklists, rollback plans, and incident playbooks that show production maturity.

Technical Defense and Review Simulation

Learn to defend your system under tough questioning with evidence, honesty, and structured reasoning that builds trust.

Week 12 Lab: Final Capstone Submission

Lock scope, finalize documentation, practice your defense, and submit a capstone that proves you can think, build, measure, explain, and defend like an AI engineer.

Unlock Full Access

Get Pro to access all chapters, lessons, and quizzes in this track.

Upgrade to Pro — $39/mo