How to Become an AI Engineer: 90-Day Roadmap

This is a realistic plan for someone starting from zero. It assumes roughly an hour a day on weekdays and three to four hours at the weekend, which is about 90 days of sustained part-time work. The most important thing about it is that every phase ends with something you can show someone.

Phase 1 — Foundations (Weeks 1–4)

• Python: syntax, data structures, functions, and object-oriented basics.
• Git and GitHub: commit, branch, and open a pull request.
• Touch SQL and REST APIs — modern AI systems are glued together with both.
• Read enough about how a model learns to be able to explain it: machine learning basics.

The temptation in this phase is to rush past it into AI quickly. Do not. Almost every AI problem you hit later is really a Python problem, a data problem, or an API problem wearing an AI costume. Four weeks here prevents two months of frustration later.

Milestone: a Python script that pulls data from an API, cleans it, and writes the result to a database, committed to GitHub with a real README.

Phase 2 — Data and ML basics (Weeks 5–8)

• Statistics essentials: mean, variance, distributions, correlation.
• Train one regression and one classification model.
• Learn to read a confusion matrix and an evaluation report.

The goal is not to become a statistician. It is to feel comfortable saying whether a model is actually any good, and to understand why a 99% accuracy number can be meaningless.

Milestone: a trained classifier with a written evaluation section, including one failure you found and explained.

Phase 3 — Language models (Weeks 9–11)

• Call an LLM API and engineer a prompt — start with prompt engineering 101.
• Understand the mechanics: how LLMs really work.
• Build a RAG system over a small document set — RAG explained simply.
• Learn the vocabulary: tokens, embeddings, context window, fine-tuning.

Milestone: a RAG app that answers questions about a set of documents you chose, where you can print the retrieved chunks and see where it goes wrong.

Phase 4 — Ship a project (Week 12+)

Pick one idea, keep it small, deploy it to a public URL, and write a one-page README. A deployed project that answers real questions is the strongest thing in your portfolio.

Write the README as if the reader is a hiring manager: what problem it solves, what you built, what you would do differently, and a link to a two-minute demo. Most candidates have projects; very few have projects that are easy to evaluate in thirty seconds.

Then build the second thing. Depth comes from repetition, not from one polished project.

Advice

• Consistency beats intensity: 45 focused minutes daily beats nine-hour weekends.
• Build in public — a short post a week compounds into a visible portfolio.
• Join a study group; you don't have to learn AI alone.
• Keep a learning log. Six months later you will not remember what you struggled with, and that is exactly what you need for interviews.
• Expect to be slow in phase 1 and fast in phase 3. That is normal and it reverses.

Go deeper

• This roadmap mirrors the Agentic AI Engineer curriculum module for module — view the track to go deeper.
• Get technical mentorship to stay accountable across the 90 days.
• Still choosing a direction? Start with career guidance.