# Embarking on the GenAI Journey: Day 01 – Demystifying the Magic

My GenAI cohort kicked off today, and what a first day! Right away we were reminded that AI isn’t magic — at its core, models like Generative Pretrained Transformers (GPT) are doing one simple thing: predicting the next token.

I jumped straight into a hands-on assignment and built a **Custom Tokenizer visualization website from scratch**.

![](https://cdn.hashnode.com/uploads/covers/5f902f13c626a0545fb6029a/db49ee80-dd9b-462c-9a7f-789ebd46b044.png align="center")

The project helped me internalize how text becomes numbers — the same idea big models rely on. Here’s a simplified look at how a query might be tokenized in this demo:

```text
Query: Hello, How are you?

Tokenization (simplified):
Hello  -> 543
How    -> 3745
Are    -> 867
you    -> 1290
?      -> 44
```

## Conceptual lessons from Day 1

Beyond the practical build, Day 1 focused on several conceptual takeaways that make the AI learning curve much less intimidating.

*   You don’t need to be a math genius to work in AI. There are many meaningful roles beyond cutting-edge research.
    
*   Understanding the ecosystem and where you fit makes it easier to choose what to learn next.
    

We sketched a simple industry split to help visualize career paths:

```text
=========================================
🔬 Machine Learning (The Researchers)
=========================================
• Focus on Research
• They create the core models
• They code the neural networks
• Heavy focus on Stats & Maths

-----------------------------------------
💻 Developers (The Builders)
-----------------------------------------
• They solve real-world problems
• They make money
• They build and deploy things
• Focus on Agentic AI & Workflows
=========================================
```

For me, knowing I’m on the “builders” side is freeing — I don’t need to rebuild models, I need to build solutions that use them.

![](https://cdn.hashnode.com/uploads/covers/5f902f13c626a0545fb6029a/61613963-ae5b-48fa-82b2-2f324f1b5a9b.png align="center")

## Vector embeddings — a quick mental model

We also learned about vector embeddings: a way to map concepts into a geometric space so related things are close together. Imagine an X/Y plot where “France” is near “Eiffel Tower” and “India” is near “India Gate.” This geometric intuition is powerful for search, recommendations, clustering, and more.

## What’s next

On my docket for upcoming posts:

1.  Explain GPT to a 5-year-old
    
2.  Explain vector embeddings to my mom
    
3.  Explain tokenization to a fresher
    

I’ll be documenting each of these as simple, shareable explainers. Follow along as I convert these lessons into approachable posts and small demos.  
  
Deep diving into a 35hrs+ course content and many other lessons.

Thanks for reading — I’ll keep sharing progress. If you’re learning GenAI too, I’d love to hear how you started and what helped you most.