Why I'm Teaching Myself Machine Learning.
A structured account of what I'm building, why I started, and the roadmap I'm following to get there.
A few months ago, I sat down and made a decision: I was going to become a machine learning engineer — and I was going to do it with a plan rigorous enough to hold up to scrutiny.
This post is my first public checkpoint. Not a highlight reel. A transparent look at where I started, what I'm building toward, and the structured system I put in place to get there.
Where it started
My first serious step into computer science was Harvard's CS50 Introduction to Computer Science. It was the right starting point — rigorous, structured, and honest about how much there is to learn. I completed it with a verified certificate, and more importantly, it reframed how I think about problems.
From there, I kept building. I completed CS50 AI with Python, learned Python and HTML, began working through JavaScript, and started shipping real projects. One of the first was Smart TaskFlow — an AI-powered productivity system built with Flask, Python, and Tailwind CSS, including a task recommendation engine. Another was Immigrant Helper, a web application co-developed to help people navigate complex administrative systems.
These weren't polished products. They were proof of concept — that I could learn, build, and ship something that worked.
The problem with learning without a system
Early on, I noticed a pattern in how most people approach self-directed learning: they follow tutorials, collect certificates, and call it a portfolio. The output looks like progress. The underlying capability often doesn't match.
I wanted something different — a structured progression that built real depth, not just surface familiarity. That led me to design ATLAS.
What ATLAS is
ATLAS is a self-directed ML Engineering roadmap I built and follow. It's organized into phases, each with clear entry criteria, exit criteria, and deliverables. There's no moving forward without demonstrating the prerequisite — the same logic a graduate program uses, applied self-directedly.
Phase 0 — Foundations: math, Python data skills, algorithms
Phase 1 — Core ML coursework + finance-focused portfolio projects
Phase 2 — Advanced ML + standardized test preparation
Phase 3 — Graduate program applications
Phase 4–5 — MS specialization + industry entry
I'm currently in Phase 0. That means building mathematical foundations (linear algebra, calculus, probability), developing solid Python data skills, and shipping real projects before moving forward. I recently completed the Kaggle Pandas certification and published a Financial EDA project analyzing AAPL, TSLA, and MSFT using Monte Carlo simulation, moving averages, and rolling volatility metrics.
Next up: Andrew Ng's Machine Learning Specialization and a deployed prediction model on Hugging Face Spaces.
Why ML Engineering specifically
The honest answer is specificity. "AI" as a career direction is too broad to be useful. ML Engineering sits at the intersection of software engineering and applied machine learning — building, deploying, and maintaining systems that actually run in production. That specificity matters for skill-building, for portfolio construction, and for knowing what to learn next.
My target niche is AI applied to finance — quantitative modeling, financial signal detection, and intelligent tooling for financial workflows. The Financial EDA project was the first concrete step in that direction.
What I'll document here
This blog will track my progress through ATLAS — technical posts, project write-ups, honest reflections on what's hard, and documentation of what I'm learning. The goal isn't to perform competence. It's to build a public record that holds me accountable and, hopefully, is useful to others doing similar work.
If you're also building in public, learning ML seriously, or thinking about what a structured self-directed path looks like — follow along. I'll be posting consistently.