Build understanding. Create something useful. Show how it works.
Learn AI engineering through a small, connected project—not a collection of unrelated tutorials. Choose a starting point, follow the selected readings beside each assignment, and advance when you can explain and reproduce the result.
| Journey | Start here when… | What you will explore | Planning budget |
|---|---|---|---|
| A: Building AI applications | You know basic Python | Software foundations, practical ML, language models, retrieval, bounded workflows, delivery | 28 weeks × 12 hours = 336 hours |
| B: Programming foundations | You are new to programming | Python, testing, data and Git, followed by A | 40 weeks × 12 hours = 480 hours |
| C: ML and data systems | You already build and deploy software | Deeper ML, data pipelines, model training, experimentation and MLOps, with applied retrieval | 32 weeks × 15 hours = 480 hours |
These are study estimates, not deadlines or guarantees. At half the weekly time, allow approximately twice as many weeks. Extend any phase when an assignment or completion gate needs more practice.
One project, growing with you
Use a learning-support workbench: a small collection of public or synthetic questions, topic labels, and licensed reference documents. Begin with a program that organizes questions. Add a classifier, answers supported by passages, and a read-only lookup workflow that produces a draft for human review.
This is an example, not a required industry or product. Another low-risk domain is fine if it supports the same exercises. Keep one domain throughout. Do not publish personal data, private documents or credentials.
Your first session
- Try A's entry check. Choose B if you need programming practice; choose C if your software foundations are solid and you want more ML/data depth.
- Write one sentence about what your project should help a person do.
- Select a small dataset you are allowed to use. Record its source, license and limitations.
- Open the first linked lesson and complete the assigned exercise. Save a short explanation and one reproducible result.
A sustainable study rhythm
For A/B, budget roughly 6 hours learning, 5 building/testing and 1 explaining/reviewing each week. C uses 6 learning, 7 implementing/evaluating and 2 reviewing. Read or watch a selected topic, try it without copying, deliberately introduce a failure, then explain the fix. Review AI-assisted code and reproduce its behavior independently; follow each course's academic-honesty rules for assessed work.
The core stack is deliberately small: Python, SQL, one API library, one database and one model interface. Named tools are worked examples, not a requirement to master every framework. Start locally; add hosted services only when a delivery exercise needs them.
A few terms, in plain language
- ML (machine learning): models learn patterns from examples; compare them with a simple non-ML baseline.
- LLM (large language model): a model that processes and generates language; fluent output can still be wrong.
- Embeddings: numerical representations used to compare items such as text passages.
- Held-out data: examples kept away from model fitting and tuning so you can test a finished choice.
- Model card: a short record of intended use, data, evaluation and known limitations.
- MLOps: practices that connect model/data versions, repeatable training, delivery, monitoring and recovery.
- Completion gate: evidence you check before advancing; not just time spent studying.
Where to go next
| Need | Open |
|---|---|
| A weekly sequence with direct resources | A: Building AI applications |
| A beginner-friendly on-ramp | B: Programming foundations |
| More ML, data and training depth | C: ML and data systems |
| Reproducible outcomes and completion gates | Project and review practice |
| A clear capability self-check | Capability guide |
| Optional operational depth after the foundations | Delivery practice |
| Course access and optional credentials | Courses and credentials |
| How the reference repository was used | Sources and verification |
| Focused extensions, not another required course | Learning resources |
Resources were selected for clear instruction, prerequisites, exercises and relevance—not a claimed universal “best” ranking. Provider and official credential names remain for attribution. The journeys do not target an employer or position, award professional experience, or promise employment.