I have spent over a decade writing production software. Now I am relearning how software works in the age of models, parameters, agents, and inference. This series captures that shift in real time.
I'm a well seasoned developer. I have 11 years of experience building and shipping web-based business applications. Today, I find myself a beginner again.
When I started experimenting with machine learning and data science in the late 2010s, I was already late. I had a lot of fun experimenting with things like scikit-learn and building small predictive models based on clustering or linear regression. But I hit a hard wall at neural nets and deep learning.
I'm not truly at ground zero. I've been a heavy ChatGPT user since 3.5 beta and have utilized various LLMs to do all sorts of cool stuff ranging from augmenting my work and development to writing custom bedtime stories for my kids involving all their favorite characters. But that's it.
Like many of us, I am stepping into a new and uncertain world where a lot of things are changing really fast and I don't know what will happen. But I know this much, I have some catching up and studying to do!
If you are already a proficient practionirer of AI tools and methodologies, you will likely find these posts to be uninteresting and redundant.
It's easier to describe what sort of questions I want to answer for myself, as these writings could benefit both individuals like myself as well as leaders who have reports that are individuals like myself who may be struggling with AI adoption and rapid change.
This is not a traditional multi-part series and does not follow a strict order. Some posts will flow naturally from concept to practice and be better read sequentially, but no strict order will be required as I will be trying to write atomically. Because of the nature of the goal, posts will naturally flow from very beginner to more advanced as I progress.
Running a LLM doesn't have to be relegated to the cloud or per-token pricing. Local models are better, our machines are more powerful, and it's easier than ever. With the right setup, you can build practical AI workflows entirely on your own hardware.