Welcome to this corner of the internet where I share my progress towards a better understanding and mastery of AI engineering. Hopefully you'll pick useful elements to go on your own journey.

The trigger

My interest in AI and data science is not new, I first had a very brief look into it in 2018 but did not go that far, favouring different things to learn.

As of August 2026, I have observed that the improvements in generative AI have made it useful enough to be here to stay. Nevertheless, the quality of its results is inconsistent, sometimes great, sometimes off topic or downright wrong. The results I hate the most are those that look plausible and feel like 99% right with a cheeky, hard-to-detect 1% wrong that ruins everything. Results can also get worse with each attempt to clarify things. The way it is prompted and the nature of our tasks seem to influence the results but is it worth the hassle?

For a few months now, coding with AI seems to have yielded good enough results for me to generate code through it but I feel that mastering the tool and what's behind it is necessary for me to build a trustworthy process around that.

I won't stop at being a power user, my end goal here is to be able to make sound choices from knowing what is behind it, turning the black box grey.

What I mean by AI engineering

By AI engineering I mean tweaking and deploying AI systems while making sound choices. It is not about developing and training models from scratch. However, understanding how it is done will hopefully help me better reason about it. The engineering activities I target are:

• choose and integrate AI models into software applications,

• architect solutions to solve specific problems,

• use mainstream tools and techniques such as RAG, Agents, and MCP,

• build an understanding of transverse topics such as security and compute costs.

I precise as well that despite AI is a vast field, I'll mainly focus on LLMs.

So… why learn?

Understand what is happening under the hood

I find it intriguing to know how those language models work and why. It might seem complex and overwhelming at first, but things are way clearer after the basics are explained.

Know what to expect from AI systems

That's the main reason why I'm diving into it. Indeed, as I said earlier, the results can be inconsistent, I have no agency over what main LLM providers deliver and prompt engineering doesn't cut it.

Be future proof

The job market has changed rapidly in less than a year. I started noticing "no code" roles popping up back in 2024, then prompt engineering, and now AI engineer and software engineer using AI.

Some roles see their tasks spill over to neighbouring ones, so we need fewer people doing them full-time — think of pure back-end developers, as fullstack ones take on part of the job.

With the current economic and political context, and the moves of big tech companies rippling across the whole world, none of us can be sure which job titles will still exist or be worth it tomorrow. That's why I'd rather add a second string to my bow.

Because that's fun too

I remember the first thing I did when discovering programming through C++ was to write a simple program approximating square and cubic roots of numbers. Just for the fun of it, by curiosity, without real-world use. That's my approach to learning about LLMs, maybe you'll enjoy it too.

What's next?

LLMs' reliability, their evaluation, benchmarks are topics I am interested in and will hopefully appear at some point. But for now, what is coming is:

• mind map of AI focusing on LLMs,

• why LLMs fail, and how prompting helps,

• integrating a model in an application: case of a job application bot.