Self-directed CS & math student. These repos are my working knowledge base: MIT and Stanford course material re-annotated bilingually (English source + 中文批注) into full notes, together with the derivation trails — where I got stuck, what I tried, the answer I finally accepted — and the agent tooling I build to maintain all of it. Only a fraction of the notes is uploaded so far. Everything here is meant to be read by both humans and AI agents — each repo has an llms.txt at its root.

Working through 6.041 (probability), 18.100B (real analysis), 6.046J (algorithms) and EE364a (convex optimization), with CS144 and MacKay's ITILA alongside. The repos below are artifacts of that pass, not a mirror of any syllabus: notes are re-derived closed-book first, then checked against the official proofs.

Three kinds of files live here, and the directory and file names say which is which:

• notes — written by me: re-derived notes, bilingual annotation, my own problem solutions and practice code. Plain text, so GitHub search and LLM crawlers can read it.
• annotated scan — a textbook page plus my handwritten derivations in the margins, scanned as an image-only PDF. No text layer: search and crawlers see only the filename. The handwriting is mine; the printed page is the publisher's.
• source — the original textbook, official handout, exam or slide deck, with no trace of me in it. A repo labelled source is a reading shelf, not a portfolio piece.

The handwritten layer is where most of the actual thinking sits, and it is the one thing a script cannot reconstruct — an OCR pass is planned so agents can read it too.

Kind follows the three-way split above. Percentages are by bytes, measured against each repo's actual file tree; where a repo is annotated scan, no percentage is given — my handwriting sits on top of the publisher's page and the two layers cannot be separated by bytes.

GitHub disallows /*/tree/ and /*/raw/ to GPTBot, ClaudeBot, PerplexityBot and friends, so directory layouts are invisible to crawlers. Each repo therefore ships an llms.txt (llmstxt.org format) stating what is really inside, with real paths.

• Docs via MCP: https://gitmcp.io/TianQue6916/<repo> — GitMCP reads llms.txt first, then README.md.
• Repo search on GitHub indexes only name, description and topics — not README body. The one-line descriptions above are written to be the search surface; the READMEs carry the detail.
• Text-layer caveat. The PDFs are mostly image-only: scanned pages, rendered slides, no OCR layer. An agent that follows llms.txt into one of those directories gets filenames, not content. Everything marked notes is plain text and is readable; everything marked annotated scan or source is not, until the OCR pass lands.

• Contents are personal study material. Textbooks and scans remain the copyright of their authors and publishers; buy the originals. On annotated scans the printed page belongs to the publisher and the handwriting in the margins is mine; bilingual notes and translations are mine.
• Repository descriptions are deliberately keyword-dense: they are the only field GitHub search actually reads.