$cat ~/posts/o-plano

bilingual · en / pt

← /blog

The Plan

2026-09-09·en·6 min read·carreiraiamlestudomanifesto

This isn't a study-tips post. I'm not going to hand you a roadmap to copy, or a list of courses with a stamp of approval. This is a record, for me, mostly. So I can come back in a year, two years, and look back and know exactly why I did what I did, and what headspace I was in when I decided.

It's called The Plan. Capitalized, because that's how it's existed in my head since August 1, 2026: it's not "a study plan," it's The Plan, the only one, full focus.

What it actually is

The Plan is a five-stage track on Coursera, running from August 2026 to June 2027, to make me a real AI/ML Engineer, not on paper, in practice. Applied NLP and LLMs, computer vision, model deployment on GPU with Docker and CUDA (something I'd never done in my life), fine-tuning, RAG, agents. The bulk of it is the IBM AI Engineering Professional Certificate (thirteen courses, PyTorch, computer vision, transformers), which replaced the Advanced Computer Vision with TensorFlow specialization I'd picked before. Swapped because it was more complete, not because it was easier.

The Plan's primary focus is foundations, not tools. But foundations alone don't ship anything, so the track also teaches the most-used technologies in the market right now, that's how PyTorch and transformers get in, for example. If something ever comes along that replaces that specific technology well without compromising the foundation underneath it, I'll swap it without a second thought.

Before this, in July, I finished Advanced NLP with spaCy. I took an NLP course on Udemy. I went back for the fundamentals again, because I didn't want to build on top of a hole.

Why this path, specifically

Deciding to "study AI" is easy, everyone is studying AI right now. The question I actually asked myself was different: why this path, in this order, and why now.

First: alignment with what I already do. I'm at Sidia, doing automated user-manual testing, and inside that, I actually apply NLP, day to day. Python is the language I use there. So the priority I chose, applied NLP/LLM plus Python for backend, is a direct extension of what I already have my hands dirty doing. I want The Plan and the job to feed each other, not run as two parallel lives.

Second: Coursera. Sidia gave me full-catalog access in July 2026, and that access expires in June 2027. That's the reason the track has a deadline. A good, globally recognized certificate you can publish on LinkedIn carries double weight for me: public weight (market, recognition) and internal weight (inside SITA, at Sidia). I wasn't going to let a window like that close without squeezing everything out of it.

Because of that deadline, actually, boot.dev (which was going to reset my Python, Linux, SQL, and Git fundamentals before the Coursera track) got cut from The Plan as I'd originally designed it. That one hurt, because I wanted that extended foundation. But doing the math was honest: there was no way to fit everything inside the Coursera window. So boot.dev didn't die, it became a parallel track, done at home, on my own time, while the mandatory specialization runs at work. I didn't give up on the foundation. I just accepted it couldn't come before anymore.

The size of it

I got guidance at work, not a rule, guidance, to not go over about 180 course hours a year, roughly 10% of annual working hours. I heard it, respected it, and decided it wasn't my limit. The full Plan, just the Coursera part, is 440 hours. I'm going to blow past the number I sent to management, and I'm going to make it more thorough along the way, not less. If I'm paying this price in hours, I want to pay it in full.

Where I am, now

I wrote the first version of this on August 17. A few days later there was already an important line to update: on the 23rd, I closed the first credential, the IBM Generative AI for Software Developers Specialization, the same one Sidia made mandatory. First credential of The Plan, delivered. And the first three requests I sent for management approval (this one, plus the ML Specialization and ML in Production) have already been approved.

In practice, I already knew a good chunk of that content, I've been using real AI in projects for a while. The course served more to put a name on what I was already doing in the dark than to teach me something new. But the credential matters in a different way: it's the first public proof The Plan is off the page.

In parallel, I got into a serious comparison of DSA courses (Stanford, deeplearning.ai, Princeton, and Augusto Galego's, which is helping me a lot right now). I identified the Stanford Algorithms Specialization as the best DSA material on Coursera. And I still want to pull the research track I've been developing at work, the paper, the possible patent, into this same structure, not as something separate.

There's a bigger reason behind all of this that I hadn't properly named in the first version: few people in my department have real interest in research. It's an open door almost nobody is trying to walk through. I want to be the one who does, I want to stand out in the "R" of R&D inside Sidia. The Plan is the technical path to that. The master's at PPGI/UFAM is the next unlock, when it's time.

Why write this down

Because a decision without a record just turns into a feeling later. A year from now I want to be able to read this and know: I didn't fall into AI because "it was trendy." I chose a specific path, in a specific order, because it grew directly out of the work I was already doing, had a real deadline closing a real window, and I was willing to pay the hour-cost that comes with it.

The Plan started on August 1, 2026. It ends, as currently designed, in June 2027. What stays on record here is who I decided to be while doing it.