A project-first ML and deep-learning curriculum for a returning learner. The repository holds the roadmap, experiments, and learning state; the AI acts as tutor, reviewer, and examiner.
Open this folder in your coding agent and say:
Start a 60-minute study session. Read AGENTS.md and my learning state, then guide me through the first diagnostic task. Let me write the learning code; give hints when I get stuck.
Start at Stage 0, even if you remember some material. The diagnostic decides what needs practice. Only Project 01 is scaffolded; create later projects when you reach them.
Use Python 3.12 and uv. From this folder:
uv sync --python 3.12
uv run python -c "import numpy, pandas, sklearn, matplotlib; print('ML environment ready')"The first sync creates .venv and uv.lock; keep the lockfile in version control after setup. This starter has dependency ranges rather than a bundled lockfile. It is an experiment workspace, so it does not install itself as a Python package. See the uv project guide.
Once you have written your first exercise file:
uv run python projects/01-linear-models/diagnostic.pyThat file is deliberately yours to create. Scripts are enough to start; add notebooks or deep-learning dependencies when a project needs them. Initial dependency installation needs internet access; the diagnostic uses a bundled dataset.
| File | Purpose |
|---|---|
| AGENTS.md | Session protocol and context routing |
| SYLLABUS.md | Stages, experiments, and exit criteria |
| Learning state | Current work, evidence, and next session |
| Concept map | Prerequisites and assessed understanding |
| First project | Scope and experiment-record format |
| Tasks | First actions and incremental challenges |
.agents/skills/ |
Tutor, experiment review, and oral exam workflows |
Say: “End this session: ask me to explain what I learned, check one prediction, and update my learning state.” Save code and experiment evidence beside the current project. Use the existing Coursera material only to answer a question encountered while building.
If your agent does not discover local skills automatically, ask it to read the matching SKILL.md listed in AGENTS.md. No custom commands are required.