[lbildzinkas]

What changed

sources.yaml gains five AI engineering sources, so coach setup downloads them next to the System Design Primer and the Tech Interview Handbook. This material feeds the AI engineering interview track (#55).

The committed list holds only material whose licence allows sharing it. Sources without a licence are kept out of the public repository: coach setup now also reads an optional, git-ignored sources.local.yaml with the same schema, for personal study on one machine.

Each source is pinned to a commit, and only the English text it needs is listed: markdown pages, no images, translations or notebooks. A fresh download of the whole committed list is 154 files, about 2.1 MB.

Source	Pinned commit	Licence at that commit	Files
alirezadir/Machine-Learning-Interviews	9ee3308	MIT	29: English guides under src/
mlabonne/llm-course	7abd96e	Apache-2.0, no NOTICE file	2: README and LICENSE
huggingface/agents-course	3c469e7	Apache-2.0, no NOTICE file	45: units/en/ fundamentals, LangGraph, agentic RAG, GAIA, function calling, observability
microsoft/generative-ai-for-beginners	d8ec07e	MIT	23: English lesson READMEs 01-21
microsoft/ai-agents-for-beginners	ff2ba66	MIT	21: English lesson READMEs 01-18 and study guide

Licences and credits

Every licence was re-checked at the pinned commit:

• The three MIT LICENSE files are the standard MIT text with their own copyright lines (Alireza Dirafzoon, 2021; Microsoft Corporation for both Microsoft courses).
• The two Apache-2.0 LICENSE files are the unmodified licence, they name no copyright holder, and neither repository has a NOTICE file.

THIRD_PARTY_NOTICES.md gives each source its own credit block: the project name and link, the authors or copyright holder as the licence or repository states them, the licence, exactly what is used, and that source's own licence notice text. The MIT blocks reproduce each LICENSE file verbatim. The Apache-2.0 blocks name the authors from the repository (the README for the LLM Course, the citation entry for the Agents Course) and carry the licence's recommended notice with a link to the full licence.

This is a reading of the licence files, not legal advice.

The optional local sources list

• load_all_sources reads sources.yaml and, when it exists, sources.local.yaml, and validates them as one list so a local source cannot reuse a committed source's name or folder.
• sources.local.yaml is in .gitignore.
• README.md explains that personal sources go in that file, are marked local_only: true and are downloaded for personal study only.

Tests

• The committed list validates, has the expected names, no translation folders, only units/en/ from the agents course, and no local-only or unlicensed source.
• The local list is added when present, is skipped when absent, cannot reuse a committed name, and is git-ignored (git check-ignore).
• coach setup downloads local sources and prints them as "(local only)".

How I checked it

• A fresh uv run coach setup downloaded all 9 committed sources (154 files, 2.1 MB, no images or translation paths). A second run printed "already up to date" for each.
• git status was clean after setup.
• scripts/check.sh passes: ruff, pyright and 39 tests.

Closes #96

[lbildzinkas]

It looks like we are not crediting people properly, I wonder whether we can really use these sources in a public repo

[lbildzinkas]

The public sources list now keeps only the five licensed sources, each with its own full credit block and licence notice in THIRD_PARTY_NOTICES.md, and the three unlicensed sources have been removed from the repository in favour of an optional git-ignored sources.local.yaml that coach setup reads for personal study.