Choose learning first; credentials are optional
Prioritize one course at a time and evidence you can reproduce. A certificate can recognize assessed learning, but it is not a substitute for a working project. These selections reflect curriculum fit and accessible instruction—not a measured prestige or employment ranking.
Provider pages checked: October 13, 2026. Access, prices, exam versions and language availability can change. Recheck before enrollment or payment.
A short, purposeful course list
| Resource | Use here | Access and completion limits |
|---|---|---|
| Harvard CS50P | B's programming on-ramp; A's testing gaps | Free OpenCourseWare; ten instructional weeks and a final project. Certificate conditions must be checked separately. |
| Harvard CS50 SQL | Selected query, relationship and schema lessons | Free OpenCourseWare; selected lessons do not complete the seven-week course and project. |
| Machine Learning Specialization — Stanford Online / DeepLearning.AI | Primary ML sequence in A/C | Three-course program. Complete required assignments for its certificate; access, subscription and assistance terms need checkout verification. Not a university degree. |
| Practical Deep Learning for Coders — fast.ai | C's selected training/deployment lessons; optional remediation in A | Free; coding experience expected. Part 1 page identifies 2022 recordings; check current library APIs. |
| Hugging Face LLM Course | A's concepts; C's tokenization, data and training | Free; good Python and introductory deep learning recommended. Selected reading does not imply course completion or a course-wide certificate. |
| MLOps Zoomcamp — DataTalks.Club | Selected tracking, deployment, monitoring and delivery topics | Free self-paced; Python, Docker and ML prerequisites. Inspected page says no 2026 live cohort and no self-paced certificate. |
The weekly pages link the actual selections; you do not need to search this table to work out what to study next. Official documentation fills specific implementation gaps. Do not buy several overlapping courses to solve one failed gate.
Understand what each evidence type means
| Evidence | Demonstrates | Does not establish |
|---|---|---|
| Course certificate | Provider's completion requirements were met | A degree, license, professional experience or mastery beyond assessed scope |
| Vendor certification | Passing the current vendor assessment | Delivery experience; coding proficiency unless actually assessed |
| Tutorial/open course | Access to instruction and practice | A credential unless the provider awards one |
| Project | Reproducible work and explained decisions | Real adoption, business impact or production experience without supporting evidence |
Optional platform credentials
Choose at most one, after project delivery, if its syllabus supports what you want to learn. None is required by A/B/C. Official credential titles are retained for accurate attribution, not as a target position.
| Official credential | Current page supports | What remains to check |
|---|---|---|
| Microsoft Certified: Azure AI Apps and Agents Developer Associate — AI-103 | Python/Foundry applications, generative/agentic features, vision, text analysis and information extraction; study guide available | Scheduling, exact version, language, local price and renewal. Practice assessment requires sign-in. The workbench alone does not cover every domain. |
| AWS Certified Machine Learning Engineer - Associate | ML workload implementation and operations. Page lists English MLA-C02 beta registration; English MLA-C01 ended September 28, 2026. Listed non-English C01 versions remain until C02 general availability. | Beta vs general availability, current guide, language, result timing, price and renewal. Do not assume a stable syllabus or immediate beta result. |
| Google Cloud Professional Machine Learning Engineer | ML/data platforms, pipelines, MLOps and conventional/generative AI; coding skill is not directly assessed | Current guide, recommended experience, exam conditions, language, price and renewal. Demonstrate Python/SQL separately. |
The Azure AI Engineer Associate page marks the certification and renewal assessment as retired. Do not plan AI-102 as a current credential or treat its older course as full AI-103 preparation.
When to pay for an exam
- Identify a specific learning purpose and the platform you can actually practice on.
- Compare the current official syllabus with your project evidence. List each uncovered domain and a focused lab.
- Confirm exam version/status, language, region, checkout price/taxes, cancellation, renewal and any sign-in requirements.
- Complete gaps and available official practice. Schedule when ready—not because the calendar reached an exam week. Scores and course completion do not guarantee a pass.
Exam study, fees and retakes are separate commitments. A/C's consolidation weeks can be substituted with exam-gap work, but do not promise complete preparation within that allocation.
Keep spending intentional
Use free course material, synthetic/local data and CPU-sized models first. Budget paid course access/certificates, hosted model usage, cloud/GPU charges, one optional exam and retakes separately. Set a monthly limit, alerts, usage limits where available and explicit teardown; a budget alert is not necessarily a hard cap.
Not verified: learner-specific discounts, checkout prices, local exam appointments, protected practice-assessment content or individual readiness. The source review confirms public page content, not an enrollment or booking transaction.