This is a personal roadmap for becoming a world-class AI Engineer.
Not a backend engineer. Not a data scientist. An AI Engineer — someone who builds AI-powered systems that work in the real world: RAG pipelines, agents, fine-tuned models, production inference, evaluation frameworks.
| File | Purpose |
|---|---|
GROWTH.md |
The full roadmap — all 5 phases, every resource, every project, every paper |
PROGRESS.md |
Your live checklist — track every item, log every hour |
| Phase | Hours | What You Build |
|---|---|---|
| 0 — Foundations | 0–500h | NumPy MLP from scratch, fast.ai, math intuition |
| 1 — The Transformer | 500–1500h | GPT from scratch (Karpathy Zero to Hero) |
| 2 — The LLM Stack | 1500–3500h | RAG systems, fine-tuning, inference, evals |
| 3 — AI Systems & Agents | 3500–6000h | Agents, multimodal, production design |
| 4 — Deep Mastery | 6000–10,000h | Papers, pre-training, original work, contributions |
- Open
GROWTH.md— read the current phase completely before starting - Open
PROGRESS.md— mark items as you go, log your hours - Build first. Read second. Write always.
- Don't move to the next phase until you can teach the current one to someone else
The best AI engineer in the world can read any paper and implement the key idea in a day. Has built things other engineers use. Is recognized by the people at the frontier.
That is what 10,000 deliberate hours builds. Inshallah.