AI Engineering Roadmap¶
A six-month, part-time path into AI/ML/LLM engineering roles — a phased plan plus an annotated library of the resources actually worth your time.
Everything here is opinionated on purpose. There is no shortage of ML reading lists; there is a shortage of guidance on what to skip. Every resource is marked with what it is and how to use it, and several well-known ones are explicitly recommended against.
Start with the roadmap :material-arrow-right: Browse the library :material-arrow-right:
The short version¶
If you have 10–15 hrs/week for six months — roughly 300 hours — this is where those hours produce offers:
| Area | Share | Why |
|---|---|---|
| DSA / coding | ~35% | The gate. Eliminates more strong ML candidates than ML does. |
| Deep learning internals | ~20% | Where applied practitioners are usually weakest. |
| ML system design | ~20% | The differentiator, and the most underprepared round. |
| ML breadth | ~10% | Necessary, but diminishing returns arrive fast. |
| Projects | ~10% | Resume-screen currency, and your interview stories. |
| Behavioral | ~5% | Cheap to prepare, expensive to fail. |
Note what's small. Math is a refresher, not a curriculum — it's the most seductive rabbit hole in ML prep, because it feels like rigorous progress while producing almost no interview value. MLOps deserves near-zero study time if you already write production software.
Kaggle competitions, MOOC completionism, and broad paper-reading are all negative ROI at this budget.
The two things that sink people¶
Preparing before applying. Pipelines run 6–12 weeks from application to onsite, so waiting until you feel ready costs a full cycle — and you will never feel ready. Apply underprepared; early interviews are calibration.
Studying the layer you already know. The comfortable material feels productive and teaches nothing an interviewer will probe. Aim at the gap, not the comfort zone.
Both are covered properly in the roadmap, including how to tell which one you're doing.
How to read the library¶
Resources are marked by intended depth:
- :material-flash: skim — know it exists, read once, move on
- :material-hammer-wrench: work through — do the exercises, write the code
- :material-bookmark: reference — return to it when you need the specific thing
Pick one "work through" resource per area and finish it. Three half-read books teach less than one completed course, and preventing that is most of why this exists.
Maintained as a working document — updated as I move through it myself. Corrections and better resources are welcome.