AI Engineer Roadmap
Python first, then building with AI agents, retrieval, and the ops around shipping models.
Four free learning paths that build on each other. Python first for eight weeks, real general-purpose Python and not data science, through syntax, functions, object orientation, and five small projects. The Vibe Coding path for six weeks on building software with AI coding agents and the judgment and security habits that separate a shipped product from a broken prototype. The AI Agents path for eight weeks on agent architectures, LangGraph, multi-agent systems, tool use through the Model Context Protocol, and retrieval. MLOps and LLMOps for six weeks on experiment tracking, model serving, evaluation, and monitoring in production.
This is the longest roadmap here at twenty eight curated weeks, because AI engineering sits on top of ordinary software engineering rather than replacing it. At ten hours a week, plan for thirty two to thirty eight weeks. At twenty hours a week, about eighteen to twenty two. Job ready is nine to fourteen months from zero, and the market rewards people who can actually deploy an agent that stays up, not people who have only run notebooks.
The order is deliberate. Python before anything AI-specific, because you cannot debug an agent framework you cannot read. Vibe Coding before the agents path, because building with AI assistance well is a skill in itself and it makes the heavier weeks faster. Agents before MLOps, because operating a thing you have never built teaches you the wrong lessons. Do the capstone in each path for real: a deployed multi-agent research assistant and a monitored pipeline are the artifacts that get interviews.
This suits people who already like building software and want to point that skill at AI systems, who are comfortable with fast-moving tools and broken documentation, and who care about the unglamorous half, evaluation, cost, latency, and failure modes. It is a poor fit if you want deep machine learning theory and model training from scratch, which is a research track, not this one, or if you want a short path, because there is no short path to this role that produces someone employable.
What this roadmap covers
Python
8 WeekLearn core Python (not data science): syntax, functions, OOP, then build real, general-purpose projects: a CLI tool, a web scraper, an automation script, a Flask app, and a simple game.
Start the free learning paths →Vibe Coding
6 WeekA free, structured path for learning to build real software with AI coding agents - Cursor, Claude Code, GitHub Copilot and autonomous agents - plus the judgment and security habits that separate shipped products from broken vibe-coded prototypes. Finish the path and pass the exam to earn a free Vibe Coding Certificate.
Start the free learning paths →AI Agents Engineering
8 WeekBuild and deploy real AI agents, not just prompt one: LangChain vs LangGraph, multi-agent systems with CrewAI, tool use via MCP, and RAG-powered retrieval, capped with your own deployed multi-agent research assistant.
Start the free learning paths →MLOps & LLMOps Engineering
6 WeekTake a model from a notebook to production: experiment tracking and a model registry with MLflow, real deployment, and LLMOps for running LLMs live, then ship a full pipeline from training to monitored production.
Start the free learning paths →The week-by-week plan
Week 1: Python Fundamentals: Syntax, Variables & Control Flow
PythonGet comfortable with Python's core syntax: variables, data types, conditionals, and loops.
Watch
- Python for Beginners - Learn Coding with Python in 1 Hour - Programming with Mosh
Milestone Set up Python and run your first script
Week 2: Functions, Data Structures & File I/O
PythonLearn to organize code into functions, work with lists/dicts, and read and write files.
Watch
- Python Tutorial for Beginners 9: Import Modules and Exploring The Standard Library - Corey Schafer
- Python Tutorial: File Objects - Reading and Writing to Files - Corey Schafer
Milestone Write and call your own functions
Week 3: Object-Oriented Python
PythonLearn classes, instances, and inheritance, the OOP foundations every real Python project uses.
Watch
- Python OOP Tutorial 1: Classes and Instances - Corey Schafer
- Python OOP Tutorial 4: Inheritance - Creating Subclasses - Corey Schafer
Milestone Define a class and create instances
Week 4: Project: Build a Command-Line Tool
PythonBuild a real CLI tool in Python using argparse to handle arguments and flags like a professional tool.
Watch
- Build a Python CLI with argparse - Alfredo Deza
- Python argparse Tutorial for Beginners 🚀 | Build CLI Like a Pro - Tech With Nanor
Milestone Structure a Python project as a runnable CLI
Project checkpoint Build a real CLI tool in Python using argparse to handle arguments and flags like a professional tool.
Week 5: Project: Build a Web Scraper
PythonScrape and extract data from a real website using requests and Beautiful Soup.
Watch
- Web Scraping with Python - Beautiful Soup Crash Course - freeCodeCamp.org
Milestone Fetch web pages with the requests library
Project checkpoint Scrape and extract data from a real website using requests and Beautiful Soup.
Week 6: Project: Automate a Real Task with Python
PythonWrite a script that automates a repetitive, real-world task: a genuinely useful automation project.
Watch
Milestone Work with the os and pathlib modules
Project checkpoint Write a script that automates a repetitive, real-world task: a genuinely useful automation project.
Week 7: Project: Build a Web App with Flask
PythonBuild and run a small full web application using the Flask framework.
Watch
- Learn Flask for Python - Full Tutorial - freeCodeCamp.org
Milestone Set up routes and views in Flask
Project checkpoint Build and run a small full web application using the Flask framework.
Week 8: Project: Build a Simple Game
PythonCap off the path by building a complete, playable game from scratch.
Watch
- Build a Tic Tac Toe Game in Python: Step-by-Step Tutorial | Python Project #24 - freeplacementcourse
Milestone Model game state and turns in code
Project checkpoint Cap off the path by building a complete, playable game from scratch.
Week 9: What Vibe Coding Actually Is
Vibe CodingBefore touching a single tool: what "vibe coding" means, where the term came from, why it works, and where it quietly falls apart if you don't know what you're doing.
Watch
- The "vibe coding" mind virus explained… - Fireship
Milestone Where the term "vibe coding" came from and what it actually describes
Week 10: AI-Native Editors: Cursor
Vibe CodingSet up and drive the editor built around AI from the ground up: multi-file edits, chat-driven refactors, and agent mode that runs terminal commands for you.
Watch
- Cursor 3.0 - Full Course for Beginners - Tech With Tim
Milestone Installing Cursor and migrating settings/extensions from VS Code
Week 11: Agentic Coding in the Terminal: Claude Code
Vibe CodingMove from an in-editor assistant to a terminal-native coding agent that can read your whole repo, plan multi-step changes, and open a working diff.
Watch
- Claude Code for Beginners Tutorial [Full Course] - freeCodeCamp.org
Milestone Installing Claude Code and connecting it to a real repository
Week 12: AI Pair Programming: GitHub Copilot
Vibe CodingThe most widely deployed AI coding assistant in production teams - inline completions, chat, and how it fits into an existing team workflow and PR process.
Watch
Milestone Setting up Copilot in VS Code and JetBrains IDEs
Week 13: Autonomous Coding Agents
Vibe CodingThe next step up from an assistant you drive: agents you hand a task to and that plan, code, test, and hand back a working result on their own.
Watch
- Devin AI Is the Future of Coding… Full Tutorial - Tech With Tim
Milestone What separates an autonomous coding agent from an in-editor assistant
Week 14: Security & Judgment: Shipping AI-Generated Code Safely
Vibe CodingThe part most vibe-coding content skips. AI-generated code ships with real vulnerabilities - this module covers the review habits and guardrails that keep a vibe-coded app from becoming a security incident.
Watch
Milestone Common vulnerability classes AI coding tools introduce silently
Project checkpoint Consolidation checkpoint: rebuild one thing from the last month from scratch with no tutorial open, then write down what you had to look up.
Week 15: What Are AI Agents? Agentic AI Foundations
AI Agents EngineeringGet the vocabulary and mental model right first. Understand what makes a system 'agentic', how agents differ from plain LLM calls, and the compound-AI architecture behind them.
Watch
- What are AI Agents? - IBM Technology
- The Power of AI Agents and Agentic AI Explained - IBM Technology
Milestone The anatomy of an agent: LLM + tools + memory + planning
Week 16: Agent Architectures & The Types of Agents
AI Agents EngineeringNot every agent is the same. Learn the standard agent taxonomy and design patterns so you can pick the right architecture for a problem instead of over-engineering.
Watch
- 5 Types of AI Agents: Autonomous Functions & Real-World Applications - IBM Technology
Milestone Simple reflex, model-based, goal-based and utility-based agents
Week 17: Choosing Your Framework: LangChain vs LangGraph
AI Agents EngineeringThe two frameworks you will be asked about. Understand where chains stop and graphs begin, and how to reason about control flow, state, and loops in agentic systems.
Watch
- LangChain vs LangGraph: A Tale of Two Frameworks - IBM Technology
Milestone Chains vs graphs: when linear pipelines break down
Week 18: Build Complex Agents with LangGraph (Hands-On)
AI Agents EngineeringThe core build module. Implement stateful, multi-step agents with LangGraph - from a basic loop all the way to a RAG-powered agent - with working Python code.
Watch
- LangGraph Complete Course for Beginners - Complex AI Agents with Python - freeCodeCamp.org
Milestone Modeling agents as graphs: nodes, edges and shared state
Week 19: Multi-Agent Systems with CrewAI
AI Agents EngineeringCoordinate teams of specialized agents. Learn roles, tasks, delegation and flows so multiple agents collaborate on a goal instead of stepping on each other.
Watch
- CrewAI Tutorial: Multiple Agents Working Together in Python - NeuralNine
- CrewAI Flows Tutorial: Master Multi-Agent Workflows - Courses by Notebook LM
Milestone Defining agents with roles, goals and backstories
Week 20: Tools & the Model Context Protocol (MCP)
AI Agents EngineeringAgents are only as useful as the tools they can call. Learn MCP - the emerging standard for connecting agents to data, APIs and applications - and ship a server end-to-end.
Watch
Milestone What MCP is and why agents need a tool standard
Project checkpoint Consolidation checkpoint: rebuild one thing from the last month from scratch with no tutorial open, then write down what you had to look up.
Week 21: RAG-Powered Agents & Retrieval
AI Agents EngineeringGive your agents grounded, up-to-date knowledge. Combine retrieval with agentic reasoning so answers are accurate, cited and current - the backbone of production agents.
Watch
- Embeddings & Vector Databases Explained - LearnThatStack
- Learn RAG From Scratch - Python AI Tutorial from a LangChain Engineer - freeCodeCamp.org
Milestone How embeddings power semantic retrieval for agents
Week 22: Project: Build and Deploy a Multi-Agent Research Assistant
AI Agents EngineeringThe capstone. Tie together LangGraph, CrewAI, tools and retrieval into one real, shippable product: a multi-agent assistant that plans research, delegates sub-tasks, and writes a grounded report.
Watch
- LangGraph Tutorial: Building a Multi-Agent AI Research Assistant - Analytics Vidhya
- Gen AI Project 2 - Build an AI Research Assistant with CrewAI | Multi-Agent System Tutorial - Siddhardhan
Milestone Designing a multi-agent graph: planner, researcher and writer roles
Project checkpoint The capstone. Tie together LangGraph, CrewAI, tools and retrieval into one real, shippable product: a multi-agent assistant that plans research, delegates sub-tasks, and writes a grounded report.
Week 23: What Is MLOps? Foundations
MLOps & LLMOps EngineeringThe third leg of the AI track: operating models in production. Understand what MLOps is, why models rot without it, and how it extends DevOps to the ML lifecycle.
Watch
- What is MLOps? - IBM Technology
- MLOps Explained - What It Is, Why You Need It and How It Works - TechWorld with Nana
Milestone What MLOps is and the problems it solves
Week 24: End-to-End MLOps Project
MLOps & LLMOps EngineeringApply DevOps principles to a real ML project. Build a production-grade pipeline with modern MLOps tooling from data to deployment.
Watch
- MLOps Course - Build Machine Learning Production Grade Projects - freeCodeCamp.org
Milestone Structuring an ML project for production
Project checkpoint Apply DevOps principles to a real ML project. Build a production-grade pipeline with modern MLOps tooling from data to deployment.
Week 25: Experiment Tracking & Model Registry with MLflow
MLOps & LLMOps EngineeringMake every model auditable and reproducible. Learn MLflow and Databricks to track parameters, metrics and artifacts, and manage models through a registry.
Watch
- Learn MLOps with MLflow and Databricks - Full Course for Machine Learning Engineers - freeCodeCamp.org
Milestone Experiment tracking with MLflow
Week 26: Deploying & Serving Models
MLOps & LLMOps EngineeringTurn a trained model into a live service. Package a model behind an API, containerize it, and deploy it to the cloud.
Watch
- How to Deploy Machine Learning Models with FastAPI, Docker, and Fly.io | End-to-End Tutorial - DataTalksClub
Milestone Wrapping a model in a FastAPI service
Week 27: LLMOps: Operating & Evaluating LLMs in Production
MLOps & LLMOps EngineeringMLOps for the LLM era. Evals, monitoring, cost and reliability for LLM-powered systems - the operational layer on top of the GenAI and Agents tracks.
Watch
Milestone Designing offline and online evals for LLM features
Week 28: Project: Ship a Full MLOps Pipeline from Training to Monitored Production
MLOps & LLMOps EngineeringThe capstone. Build one real, named MLOps pipeline start to finish: scrape and clean data, track experiments with MLflow, version a DVC pipeline, deploy with Flask and Docker, and automate releases with CI/CD.
Watch
- MLOps Pipeline with Python, AWS, Docker - YouTube Viewer Sentiment - freeCodeCamp.org
Milestone Planning a full pipeline architecture before writing code
Project checkpoint The capstone. Build one real, named MLOps pipeline start to finish: scrape and clean data, track experiments with MLflow, version a DVC pipeline, deploy with Flask and Docker, and automate releases with CI/CD.
Resources and structure are drawn from the free Dropout Developer learning paths.
Get this as a dated plan
Run the same roadmap through the generator with your weekly hours and a target date to get a dated schedule you can copy, download, or print.
FAQ
Do I need a machine learning or maths background?
Not for AI engineering as this roadmap defines it, which is building and operating systems on top of existing models. You need solid software engineering, comfort with APIs, and an understanding of evaluation and cost. Deep ML theory is a separate research track.
Is "vibe coding" a real skill or a meme?
Both. Letting an agent write code with no judgment produces prototypes that break. Using AI assistance deliberately, with review habits and security guardrails, is a genuine productivity skill and is why that path is early in this roadmap.
How fast is this field changing, and does that make the roadmap pointless?
Tools change monthly, the shape does not. Agents, retrieval, tool use, evaluation, and deployment have been the structure for a while. Learn the structure and swapping a framework is a weekend, not a restart.
What should my portfolio show for an AI engineer role?
One deployed multi-agent or RAG system with a public URL, an evaluation harness, and a short writeup of what failed and how you measured it. The writeup matters as much as the demo, because it shows you understand production, not just prompts.
Can I do this alongside a full-time job?
Yes, at ten hours a week, expecting roughly nine months. The Python and Vibe Coding weeks are the most job-compatible. The agents and MLOps capstones need a few uninterrupted weekend blocks to finish properly.