How to Start Learning AI in Tanzania (or Anywhere in Africa) in 2026

A friend in Dar es Salaam messaged me last month. He's twenty-three, finishing a finance degree, and the only thing he wants to do for the next decade is build with artificial intelligence. He had one question: where do I start?

I've answered some version of that question fifty times in the past year. Every answer was bespoke. This essay is the version I would write if I could only write it once.

This is not a list of YouTube videos. This is a roadmap.

The hard truth: most courses are not the bottleneck

People assume the hard part of learning AI is finding the right course. It isn't. There are thousands of free, world-class resources online — Andrew Ng's, fast.ai, Hugging Face's, MIT's. The hard part is the boring part: showing up every day for eighteen months while your friends are doing other things.

So before we discuss curriculum, internalize this: the constraint is not access. The constraint is consistency. A learner in Mwanza with a cheap laptop and a stubborn habit will outperform a learner in Nairobi with the best tutors and an inconsistent schedule.

Phase 1 — The foundation (months 1–3)

Forget about transformers. Forget about ChatGPT. The first ninety days are about three things, in this order:

1. Python, fluently

Not "I can read Python." Fluent — you can sit down and write a 200-line script without Googling syntax. Use Python's official tutorial, then Automate the Boring Stuff. Build small things: a script that renames your files, a scraper that pulls headlines from a news site, a tiny calculator. Boring is the point.

2. The mathematics you actually need

You do not need a math degree. You need:

• Linear algebra basics — vectors, matrices, dot products. Watch 3Blue1Brown's Essence of Linear Algebra (free, on YouTube).
• Calculus, conceptually — derivatives and gradients. Same channel, "Essence of Calculus."
• Probability and statistics — distributions, expectation, Bayes' theorem.

Don't grind exercises. Watch the videos, take notes by hand, and move on. You'll come back when you need it.

3. The command line, Git, and how the internet works

This is the part everyone skips and everyone regrets. Spend two weeks learning bash, Git, and HTTP. You will use these every single day for the rest of your career.

Phase 2 — Machine learning fundamentals (months 4–7)

Now we enter the real work. Two courses, in order:

The two non-negotiable courses

• Andrew Ng — Machine Learning Specialization (Coursera)

  Free

  The classic. Audit it free; do every assignment. Strong foundation in supervised learning, gradient descent, regularization.

  Open

• fast.ai — Practical Deep Learning for Coders

  Free

  Top-down, build-first approach. By the end of lesson 4 you'll have trained a real model.

  Open

Some links may be affiliate. We only recommend tools we have personally vetted.

A common mistake: people start fast.ai first because it's faster and more exciting. Don't. Andrew Ng builds the intuition that lets fast.ai's magic actually make sense. Order matters.

While you're working through these, publish what you learn. A weekly blog post, a tweet thread, a small repo. The act of explaining cements the knowledge — and starts your public track record, which will matter more than you think when you start applying for jobs.

Phase 3 — Modern LLMs and the AI stack (months 8–12)

By month eight, you'll know enough to engage seriously with the current state of the field. Focus on:

Reading the foundational papers

Read these, in order, with a pen. Don't try to understand everything. Try to understand the shape:

1. Attention Is All You Need (Vaswani et al., 2017) — the original transformer.
2. BERT (2018) and GPT-3 (2020).
3. InstructGPT (2022) — how RLHF works.
4. Sparks of Artificial General Intelligence (Bubeck et al., 2023) — the GPT-4 capabilities paper.

Building real things

Pick a project. Anything. A study buddy chatbot, a tool that summarizes Swahili news in English, a code reviewer for your own GitHub commits. The project is the curriculum at this stage.

Tools to learn deeply

Hugging Face Transformers. LangChain or LlamaIndex (one of them, not both, this year). The OpenAI and Anthropic SDKs. Vector databases — pgvector is enough; you don't need anything fancier.

Phase 4 — Specialize, ship, and be seen (months 13–18)

You now know more than 99% of people who claim to "know AI." Pick a direction:

• Research-leaning: apply for residencies, contribute to open-source models.
• Engineering-leaning: ship a small SaaS, freelance on real projects, or join an early-stage team.
• Product-leaning: build an AI-augmented business in a vertical you understand — agriculture, finance, education, health.

For most readers in Africa, I recommend the product path. The frontier is not lacking for researchers. It is lacking for people who can take models and put them in the hands of farmers, traders, students, and small businesses. That work pays well, compounds quickly, and is genuinely needed.

The African reality — and the African advantage

Yes, electricity is unreliable. Yes, bandwidth is expensive. Yes, the local AI community is small. These are real constraints.

But here's what most articles about "learning AI in Africa" miss: you have advantages too. The same models that cost $10,000 of compute to train two years ago can now be fine-tuned for free in a Colab notebook. The tools have gotten radically better and radically cheaper, and they reached you at the same moment they reached Stanford. The asymmetry of access — for the first time in computing history — is closing.

You are not behind. You are early.

A practical starter kit

If you're starting today, here is the minimum:

• A laptop with at least 8GB of RAM. It does not need a GPU; Colab gives you free ones.
• A reliable, daily two-hour block. Same time, same place.
• A folder on GitHub where everything you build lives — public, even if it's bad.
• A way to ask questions: a Discord, a study group, or this site's comments.
• One specific problem in your environment that you want AI to help solve.

Start there. Eighteen months from now, message me.

Further reading on this site

• What is Cybersecurity? A Plain-English Guide — because every AI engineer should understand the systems they build for.
• Building a Personal Brand as an African Technologist — your work matters less than people knowing it exists.
• Browse the Technology category for more.

If this essay was useful, subscribe to the newsletter. Sundays only, one essay, no fluff.