← Back to Articles

Introduction

A brief introduction to this series of articles to tell my journey towards AI software development.

I come from Java/JavaScript development. Those last few years I have developed using Node.js and React.js mostly. In the last year I started working on powering my English teaching platform with AI. I am currently working on my software development skills and exploring the new possibilities with AI.

I am interested in all aspects, ranging from local video generation to optimised agentic workflows to tress placement appreciation in speech.

I would like to start by saying a few words about some of the most important concepts and techniques that I am endevouring to learn about as they seem to be of the utmost importance in today's AI landscape.

LLMs

Large language models or LLMs and especially the possibility of running them locally represents a huge potential. There is no denying that the multiplication of AI providers and infinite possibilities and combinations can sum up to huge expenses that make you take tapes towards independence.

I have started testing a few LLM's locally using LM Studio or Ollama.

LLMs and VS code

I have started implementing an integration to VS code in the chat, the Codex and the Continue panels. So far I've had the best result via the continue panel. The integration of the autocomplete happened seamlessly. I especially recommend Kyle's video I will be linking below.

Regarding VS code integration with VS code, I have had the best result via the continue panel. The LLMs I have tested as of today 9/09/2026 are:

Setup Model Endpoint VS Code integration Tools Result Current verdict
LM Studio qwen/qwen3.8-27b 127.0.0.1:1234 GitHub Copilot / customoai true Appeared, but extremely slow / fans ❌ Not practical
LM Studio qwen/qwen3.8-27b 127.0.0.1:1234 GitHub Copilot / customendpoint true Sometimes appeared, then hung ❌ Not practical
LM Studio qwen/qwen3-4b-2507 127.0.0.1:1234 GitHub Copilot / customendpoint true Raw API worked; Agent still problematic ❌ Copilot path problematic
LM Studio qwen/qwen3-4b-2507 127.0.0.1:1234/v1 Qwen Code Companion N/A LM STUDIO WORKS via API ✅ Endpoint works
Ollama qwen2:7b 127.0.0.1:11434/v1 GitHub Copilot / customendpoint true Model appeared, but Agent hung ❌ Copilot path
Ollama qwen2:7b 127.0.0.1:11434/v1 Qwen Code Companion 0.21.11 native agent tools Agent eventually started; permissions worked 🟡 Works partially
Ollama qwen2:7b 127.0.0.1:11434/v1 Qwen Code Companion 0.21.11 filesystem create hello.txt worked 🟢 Tooling partly works
Ollama qwen2:7b 127.0.0.1:11434/v1 Qwen Code Companion 0.21.11 read/edit Hangs; Ollama sees nothing 🔴 Companion/ACP problem
LM Studio qwen/qwen3.8-27b 127.0.0.1:1234 Continue Agent-ish / coding Works! 🟢 Best result so far

Hugging Face

Hugging Face is at the moment the hub that enables you to best comprehend the extent of development of LLMs. Structure enables you to test and study LLMs and structure and gives it the best visibility. I discovered hugging face thanks to a few videos on YouTube among which one by Kyle from Web dev simplified that explained how to run LLMs locally.

I am following the AI agent free course On Hugging Face. I recommend it because hogging face hugging face is definitely the hub where all AI tech people should display their work. Having said that I find that the course can be difficult to follow at times because it references

Python

It's been a few attempts for me to try and learn python and it has been extremely frustrating because so far the experience was extremely boring because most tutorials are addressed to complete beginners. I have tried on several occasions to follow tutorials and failed miserably out of boredom. Lately I have convinced myself that I needed to make another attempt because Python has become unavoidable for anyone who wants to progress in AI.

One of the main issues has been the widespread use of "sandbox" environments like Jupyter or consoles that unnecessarity and ridiculously complicate the learning experience. I much rather use an IDE like VS code the same way I have learnt to code in all the programming languages I have learnt so far.

finally made up my mind to give it another chance and look for tutorials aimed at Java or JavaScript developers. I am extremely satisfied with the course I found on Udemy and I will share the link at the end of this article. The course is called "Become an AI Engineer: Python for JavaScript Developers" by Vilva Athiban. It is very well structured and it is a great way to learn Python if you are already familiar with JavaScript. It is not free but it is worth every penny.