I know that mobile game developers need to make money but the ad networks they use are the worst. They put a >> or a X in the corner and you have to press that button a dozen times for the ad to eventually close. That’s quite scammy. I’ve been thinking of introducing ads in my apps and this just feels wrong and greedy.
I’ve been reading quite a lot around AI safety, and one thing I’ve found frustrating is how difficult it is to find material that bridges the gap between talking about AI safety and actually doing AI safety research.
There are plenty of good books and papers discussing the political, socioeconomic, philosophical and governance questions around AI. Those conversations are obviously important but if you’re a data scientist, ML engineer or someone already working with AI few tell you how doing AI safety research actually looks like in Python?
For example how to actually run an experiment looking at sycophancy, how to create or use a dataset to test whether a model changes its answer because of information about the user, how to measure the behaviour, what does the evaluation code look like, how do you interpret the results, and what can you not conclude from them?
That is the gap I’m trying to address with a book I’m currently writing.
The approach is very practical. Each topic starts with the safety problem and the research behind it, but then we actually build the experiment. Python code, datasets, models, metrics, results and discussion of the limitations, all those.
The idea isn’t to pretend that running a few notebooks suddenly makes someone an AI safety researcher. It’s to make the field much more approachable to people who already have data science or AI skills and want to understand what technical AI safety research actually involves.
I wrote a technical book on Practical LLMOps and this was one of the chapters. Because of the sheer amount of content, I stripped it bare in that book and made it its own.
I’ve just published Chapter 1 on Substack (just finished chapter 4). It introduces the approach I’m taking with the book and starts building that bridge between AI safety as a subject people discuss and AI safety as something we can actually investigate experimentally.
Would genuinely be interested in feedback, particularly from people already working in ML, data science or AI safety.
https://open.substack.com/pub/houstonmuzamhindo/p/i-am-writing-a-practical-technical
A little while ago I posted here asking about interest in practical AI learning for Zimbabweans. Thank you to everyone who engaged and provided feedback! The response was better than I expected. I had people messaging me here, on WhatsApp and elsewhere asking about the classes, what they would cover, whether you needed to know how to code, and when we would start.
So I decided to actually build it. I have now called it called “Applied AI School.”
The idea originally started with Zimbabwe, but after I began talking about it publicly, I started getting messages from people who aren’t Zimbabwean asking if they could join too. So I’ve built it in a way that can grow beyond Zim while keeping the original goal: making high-quality, practical AI education accessible without people needing to spend hundreds or thousands of dollars.
The first live course starts Saturday, 5 September
Level 1: AI Foundations – FREE – 4 weekends
This is for anyone who wants to understand and use AI properly. No coding or previous AI experience required.
We’ll cover how modern AI actually works at a useful level, prompting, tokens and context, hallucinations, where AI is useful and where it isn’t, responsible use, and how to apply it to real work, study, business and everyday problems.
The course booklet is already available. Once you create an account and enrol, you’ll find it under the Resources tab so you can start reading before the first live session.
Level 2: Applied AI Development – 8 weekends
This is where we move from using AI to actually building with it.
We’ll start with the Python you need and work with LLM APIs before moving into embeddings, vector search and RAG, giving models access to tools, and building agentic workflows.
The emphasis will be on actually building things rather than just learning the terminology. By the end, you should understand how the pieces behind modern AI applications fit together and be able to build your own.
Level 3: Advanced LLM Engineering / LLMOps – 10 weekends
This is the deeper technical track for people who want to understand what happens underneath the APIs and how LLM systems are actually run in production.
We’ll go into areas such as tokenisation, inference and the prefill/decode process, KV caching, batching and scheduling, quantisation, speculative decoding, evaluation and observability, deployment, scaling and the cost and reliability considerations involved in running LLM systems.
So there is a pathway from never having touched AI before, to using it properly, to building AI applications, and eventually understanding the engineering required to deploy these systems.
All of the cohort classes are taught live on Google Meet, so you’ll be able to ask questions and work through things with us rather than simply watching recordings.
There are also self-paced courses you can take in your own time (Supervised Machine Learning from First Principles, Unsupervised Machine Learning from First Principles, Building a Large Language Model from First Principles, and Building a Reasoning Model from First Principles).
And we’ve got a free Applied AI School WhatsApp Community for discussions, questions, resources and updates. The link to join is in the footer of the website.
If you were interested when I first posted, you can create an account and enrol now:
For anyone starting from scratch, enrol in the free AI Foundations course and I’ll see you live on Saturday, 5 September.
It feels good to finally come back here with something people can actually use after that original post 😅. As always, feedback is welcome.
Last week, I asked whether people would be interested in affordable, practical AI courses created with Zimbabweans in mind.
The response in the comments and on my social media statuses was encouraging. I also heard from people outside Zimbabwe, while others were interested but did not identify as “AI engineers.”
Initially I was going to call it AI Engineers Zimbabwe but now I have settled on Applied AI School, a broader school for anyone who wants to understand, build with or engineer AI.
The school will have two learning pillars:
1. Live learning
Structured courses taught live on Google Meet, with exercises, projects, recordings and lifetime access to the course material.
2. Free self-paced learning
I am adding courses I have taught over the years for free for anyone to study in their own time, without joining a live cohort. These are:
The platform has not launched yet and I hope to share the website and dates for the first live course soon (planning for 1st of September for the first lesson). Thank you to everyone whose feedback helped give the idea a much clearer direction. :)
Ndeip guys.
I’m Zimbabwean, currently based in London and I’ve been thinking about running a live online AI Engineering course specifically for people back home and in the diaspora.
I’ve been following some of the AI and tech discussions on this subreddit for a while. I’ve seen people asking about learning Python and AI, getting beyond the theory taught at university, what skills are actually useful for getting into tech, and generally how Zimbabweans can position themselves as AI becomes more important.
Seeing some of those discussions is actually part of what got me thinking about doing this.
Before I build the whole thing though, I wanted to see if there is genuinely enough interest here.
One thing I don’t want this to be is another course showing people how to use ChatGPT, Claude or Copilot. There is already plenty of that content online for free.
I want to teach the harder stuff: how modern AI systems actually work and how you build them.
Things like Python for AI, machine learning foundations, transformers and attention, embeddings, RAG, vector databases, agents, fine-tuning, evaluation, deploying models, and eventually some of the deeper LLMOps topics like inference, KV caching, quantisation, batching and serving models.
The idea would be to build up to these concepts rather than assuming everyone already knows them, with practical projects throughout. I want people to finish having actually built things and understanding what they built, not just having watched some Zoom calls and received a certificate.
A bit about me so you know who’s teaching it:
I’m a Zimbabwean AI/data science professional with around 8 years of experience working in the field. I’m currently Head of Applied AI at a financial services organisation in the UK, where I’ve worked on taking AI systems from ideas through to production.
I’ve also taught AI and machine learning online to thousands of students through Udemy and my own education platform.
I recently wrote and self-published a 300+ page technical book, Practical LLMOps Deep Dive: Building Production LLM Systems from First Principles, which goes into things like transformer inference, KV caches, batching and scheduling, quantisation, deployment, observability and evaluation.
I worked for OpenAI part time during the training of the GPT-4 model and for exposure to how top labs do certain things like RLHF.
Outside of my day job, I’m a member of the UK Finance AI Policy Committee, where I contribute to discussions around AI in financial services.
I’ve also had the opportunity to speak on AI in Zimbabwe as part of a UN/Government of Zim discussion, particularly around practical applications of AI across government, education, industry and other areas.
So this isn’t something I’m teaching because AI suddenly became popular. It’s the field I’ve been working in for years.
Why Zimbabwe?
There is clearly going to be increasing demand for AI and digital skills around the world, including back home. I would love to see more Zimbabweans developing the skills to actually build this technology, rather than us only becoming consumers of tools built elsewhere.
I’ve been fortunate enough to build my career in this space, work on real AI systems and learn a lot along the way. If I can take some of that knowledge and make it accessible to people back home who genuinely want to get into the field, I think that’s worth doing.
I also know the economic reality in Zimbabwe, so I don’t want to take something priced for someone earning a UK or US salary and simply market it back home. Whatever I put together needs to be realistically accessible.
But I don’t want “accessible” to mean watered down either.
If we’re going to learn transformers, we’ll learn transformers. If we’re going to learn RAG, we’ll actually build one. If we get to quantisation, we’ll understand what we’re quantising and why.
The standard should still be high.
It would be live rather than purely prerecorded, probably evenings/weekends, with recordings available for anyone who has connectivity problems.
I’m still designing it, so rather than deciding what Zimbabweans need from London, I’d genuinely like to hear from people back home.
Would you be interested?
If yes, tell me a little about where you’re currently at, whether you’re a student, graduate, developer, already working in tech, or completely new to this and what parts of AI you’d most like to understand.
If there’s enough interest, I’ll put together the first cohort and we’ll see where it goes.