7 free courses to learn software and AI from scratch, in order
There are hundreds of courses and nearly all of them start with the same thing. These seven go in order: each one starts where the previous one ends, none repeats another, and all seven can be done for free. They come from comparing around 300 pages of guides, roadmaps and forums in Spanish and English. All seven courses are fully free; the only paid things are two optional certificates (edX's for CS50 and PortSwigger's for security), the API key the last course's exercises need, and Google Cloud in the Kubernetes one, which the $300 credit for new accounts covers.
At two hours a day, five days a week, doing all seven takes 12 to 18 months (520 to 770 hours). The hours for CS50, Full Stack Open and DevOps with Kubernetes come from their websites; for the other four they're my estimate based on how long the material runs.
| Step | Course | What you learn | Hours | Free | Spanish |
|---|---|---|---|---|---|
| 1 |  CS50 | Real programming: C, Python, SQL and web | 110-220 | Yes, with a free certificate | Dubbed lectures |
| 2 |  Full Stack Open | Building and deploying a web app with React and Node | 120-160 | Yes, with a free certificate | Almost all of it |
| 3 |  Web Security Academy | Attacking and securing web apps: SQL injection, XSS, CSRF and more | 60-100 | Yes, no certificate | No |
| 4 |  DevOps with Kubernetes | Putting apps in the cloud with Kubernetes, CI/CD and scaling | 100 | Yes, with a free certificate | No |
| 5 |  Amazon MLU | Classic machine learning, with interactive visuals | 20-30 | Yes, no certificate | No |
| 6 |  fast.ai | Deep learning: training and deploying models | 80-100 | Yes, no certificate | No |
| 7 |  Generative AI for Beginners | Building with LLMs: prompts, RAG and agents | 30-60 | Yes, no certificate | Yes |
1. CS50, to learn to code (110-220 hours)
Harvard's intro to computer science, and the first step in nearly every roadmap I looked at. It's 11 weeks: you start with Scratch, move to C to understand what memory does underneath, then algorithms and data structures, Python, SQL, HTML, CSS and JavaScript, and a web app with Flask. There's also a lecture on AI (neural networks, LLMs, hallucinations), and you finish with a project of your own.
It's free. If you score 70% on every problem set and submit the final project, Harvard gives you its certificate for free; the verified edX certificate costs $219. Lectures are dubbed into Spanish, it adds up to 110 to 220 hours (10 to 20 for each of the 11 weeks), and the current edition closes on June 30, 2027.
Weeks 4 and 5, on memory in C, are where most people drop out. If you get stuck there, ask the duck at CS50.ai, the course's own AI tutor, before giving up. cs50.harvard.edu/x ↗
2. Full Stack Open, to build web apps (120-160 hours)
The University of Helsinki's course on building modern web apps. CS50 teaches you to code, and here you build an app like the real ones: React for what users see, Node and Express for the server, databases, tests, TypeScript, GraphQL, continuous integration and containers. By part 3 you're already putting it online with Fly.io or Render.
It's free and comes with a certificate and ECTS credits from the University of Helsinki. Each part takes 15 to 20 hours, so the eight main ones (0 to 7) add up to 120 to 160, and almost all of it is translated into Spanish. It assumes you can already code, which is exactly what CS50 leaves you with. fullstackopen.com ↗
3. Web Security Academy, so your app doesn't get hacked (60-100 hours)
The free academy from PortSwigger, the company behind Burp Suite, the tool pentesters use to audit websites. Each topic explains one kind of attack and then gives you labs, real vulnerable sites you have to break yourself: SQL injection (16 labs), XSS (30), CSRF (8), access control, authentication, OAuth, JWT, SSRF, file uploads, API and GraphQL attacks, and more recently attacks on LLM-powered apps. There are over 30 topics, and new ones get added as new attacks appear.
It's 100% free and always open: you create an account and it tracks your progress. The labs are done with Burp Suite Community, which is also free. There's no certificate; theirs (Burp Suite Certified Practitioner) is paid and optional. It's English only and doesn't give hours, so the 60-100 are my estimate for the beginner and intermediate topics. It comes after Full Stack Open because you already know how a web app is built and here you learn to break it, and before putting it in the cloud. portswigger.net/web-security ↗
4. DevOps with Kubernetes, to put your apps in the cloud and scale them (100 hours)
The next University of Helsinki course, run since 2024 by Matti Luukkainen, the same teacher behind Full Stack Open. You set up a Kubernetes cluster on your own machine with k3d and deploy your apps to it with deployments, services, ingress and gateway. Then you move them to Google Kubernetes Engine, make them scale on their own with load, monitor them with Prometheus and Grafana and automate deployments with GitOps. It ends with service meshes, custom Kubernetes resources (CRDs) and how it all works underneath.
It's free and comes with a certificate and 5 ECTS credits if you submit at least 30 of its 55 exercises; the grade goes up with how many you do. The course itself estimates about 100 hours, it's English only, and the current edition closes on January 31, 2027. It asks for Docker and web development, which is exactly what Full Stack Open and its containers part leave you with. Google Kubernetes Engine is paid, but new Google Cloud accounts get $300 in credit, and the rest of the course runs on your own machine. devopswithkubernetes.com ↗
5. Amazon's Machine Learning University (20-30 hours)
What Amazon used to train its own engineers, open to everyone. It has three pieces that fit together. MLU Explain is 14 interactive articles where you move the data and watch what happens to the model: linear and logistic regression, decision trees, random forests, neural networks, precision and recall, the ROC curve and the bias-variance tradeoff. On YouTube there's the «Machine Learning through Application» course (17 videos, almost 4 hours), and on GitHub the notebooks with exercises and a final project.
It's free, but there's no certificate and it's English only. You need Python, since the exercises use scikit-learn, and the notebooks run for free on SageMaker Studio Lab or Google Colab. Watch out for the 2022 press release that still circulates: the YouTube courses it names no longer exist, and the current ones are on the official channel. Between videos, articles and notebooks I estimate 20 to 30 hours. MLU Explain ↗ · YouTube channel ↗
6. fast.ai, for deep learning (80-100 hours)
Practical Deep Learning for Coders works the other way round from most courses: first you train a model that works, then you learn why. Over 9 lessons of about 90 minutes you build image, text and tabular models with PyTorch, and publish them for others to use. MLU explains the concepts with visuals and here you code them, so they don't overlap.
It's free, with no certificate, in English with captions. It asks for about a year of coding and high-school math; it explains the rest along the way, and the notebooks run for free on Google Colab or Kaggle without a graphics card of your own. It's 13.5 hours of video, and counting the practice for each lesson I estimate 80 to 100 hours. course.fast.ai ↗
7. Generative AI for Beginners, to build with AI (30-60 hours)
Microsoft's course on GitHub for building apps with language models. Its 21 lessons go from your first prompt to what's used in production: calling functions from the model, RAG with vector databases so it answers from your documents, and agents. The code is in Python and TypeScript, and it doesn't tie you to any framework.
The course is free and translated into Spanish, but there's no certificate and the exercises need an OpenAI or Azure API key, which is billed by usage. If you want a certificate afterwards, Hugging Face's agents course is free and gives one. I estimate 30 to 60 hours, depending on how long you spend on the exercises. github.com/microsoft/generative-ai-for-beginners ↗
And the math
It doesn't get a slot because none of these courses asks for more than algebra and some statistics, and fast.ai teaches the rest along the way. For steps 5 and 6, the 3Blue1Brown neural network series, which goes up to transformers, takes an afternoon and makes everything much clearer. If you're missing the basics, Khan Academy covers them for free.