A documentation of the AI learning journey, built in public. Each lesson is a real step taken, written down as it happened.
Learning sticks when something real comes out of it. These are the projects woven into the path.
An AI that reads PDFs and notes, then answers questions from them with precision - built using RAG and embeddings.
An agent that doesn't just answer - it uses tools, makes decisions, and takes actions to reach a goal on its own.
A full AI companion with memory, retrieval, and personalization - the project where everything on this path comes together.
Not the hype, not the fear. A clear-eyed look at what AI actually is - and what it means to build with it.
Peek inside the mind of an LLM using stories, not formulas.
Once you understand tokens, strange AI behaviour stops being mysterious - and starts being predictable.
The same question, asked two different ways, produces completely different answers. That gap is what this lesson is about.
There is a moment when you stop reading about AI and start talking to it in code. This is that moment.
A thoughtful answer is wonderful. An answer shaped exactly the way your application needs - is something else entirely.
Meaning, turned into numbers. Distance, used as similarity. One of the most elegant ideas in modern AI.
An LLM only knows what it was trained on - nothing more. RAG is how you change that, and you will use it in almost every real project.
Every conversation resets. What does it take to build something that doesn't forget? More than you'd think - and less than you'd fear.
The patterns you have been learning - chains, retrievers, memory - are already built. This is how you use them with real understanding.
Language models describe the world. Tool-using models act in it. This lesson is about the shift between the two.
Rushing to an answer is a human flaw, and an AI one too. Teaching your model to slow down changes what it is capable of.
Until now, you have been writing prompts. This is when you start writing goals - and the model figures out the path.
Intuition tells you something feels right. Evaluation tells you whether it actually is. You need both - but most developers only have one.
Everything you have learned lives in your head until something real exists. Today, that changes.
Your AI is live. Something feels off. Without the right tools, you are guessing. This lesson ends the guessing.
Not an exercise. A real project, from first principles, built entirely by you over five days.
Before Vaarta can speak with any depth, it needs to know things. Today you build the layer it reasons from.
Fetching the right information is half the work. Knowing what to do with it - is the harder, more important half.
A system that remembers who you are is a different thing from one that doesn't. This is how Vaarta becomes that.
There is something clarifying about shipping. Your work becomes real the moment someone else uses it.
Prompting changes how AI speaks. Fine-tuning changes who it is. Knowing which one you need is the real skill.
The world is not made of text. This lesson is about building AI that meets it more fully.
Speed and cost are not afterthoughts. They are design decisions - and they shape what you can actually afford to build.
A project that runs only on your laptop is an experiment. Getting it in front of someone else is the real test.
No API key. No rate limits. No dependency on anyone else. What that independence actually looks and feels like.
A billion people deserve AI that speaks their language. This is one of the most important problems left to build.
The most important question about any AI system is not what it can do - it is what happens when it doesn't.
The tools you have been learning were built by people who once asked the same questions you did. This is how you join them.
Thirty days in, and something has shifted. This is where the notes end and the real building begins.
New notes are added as the journey progresses. Subscribe to stay in the loop.
Subscribe for Updates