As someone working in software development, primarily focused on building web products, I’ve always been curious about emerging technologies. The explosion of interest in AI, particularly after the release of ChatGPT, sparked my desire to dive deeper into this fascinating field. Here’s how my journey unfolded.
I started with YouTube videos inlcuding (Wolfram’s explanation) and Deep Dive into LLMs like ChatGPT that breaks down how ChatGPT predicts the next word in a sentence (if you don't want to watch the video, you can read the transcript in the article).
This introduction was an eye-opener—it revealed just how advanced AI, particularly large language models (LLMs), had become. It also introduced me to foundational concepts in the LLM space, which set the stage for my deeper exploration.
From there, I began learning about different LLMs such as Llama, GPT, Claude, and others. This phase was all about understanding technical aspects like context length, tokens (input and output), parameters, embeddings, FLOPS (Floating Point Operations Per Second), RAG (Retrieval-Augmented Generation). To make things more practical, I experimented with running LLMs locally using tools like Ollama. It was my first hands-on experience with local models, and I was amazed at their potential.
Some resource during this phase:
- All you need to know about Tokenization in LLMs
- Llama Recipes Quickstart
- Everything I've learned so far about running local LLMs
- Understanding LLMs from Scratch Using Middle School Math
Understanding those tech terms are not enough, we have to explore and get hands dirty to truely understand how these terms are interconnected and where they play a role in and why they are needed. I explored them using Jupyter Notebook—a tool that allowed me to run Python instructions for various tasks. This led me to platforms like LangChain and Kaggle, where I experimented with basic LLM training and model fine tunings. These experiences helped me understand how AI models are trained and fine-tuned, giving me a more comprehensive view of their capabilities.
Some resource during this phase:
Armed with this growing knowledge, I started building AI-powered applications (Sudoku, Page Saver) using tools like Cursor and Windsurf, which are enhanced by AI-driven editors. I quickly realized how transformative AI can be for developers. Many challenges I faced in software development could be streamlined or even entirely solved by AI. This was a pivotal moment that underscored AI’s immense potential to reshape the way we work.
The journey continues as I really started to build something with prompting, first begun with some examples of calling different APIs using different frameworks. Since I am primarily using GoLang at the moment, so I praticed with GoLang. Some tries are recorded into some blog posts like:
- First Touch on Open API
- Gemini Example with Go
- Calculating Token Count for Claude API Using Go: A Step-by-Step Guide
With all these basics, a more complex application powered by AI can be built.
While my journey is still ongoing, one thing is clear—AI is not just a trend: it’s the future. Each step I’ve taken has deepened my appreciation for the field and its possibilities. I’m excited to see where this journey takes me next and how AI will continue to revolutionize our lives, both as developers and as individuals.
great article