Machine Learning Systems Reference Plugin

A searchable study and reference package derived from Vijay Janapa Reddi's open textbook, Machine Learning Systems. It exposes the material as Markdown and through a small FastMCP server.

This repository is an independently organized derivative reference. It is not the original textbook, an official Harvard project, or an operational MLOps platform.

Contents

• 44 chapter and appendix files across two volumes (Volume 2 has no Chapter 4 in this collection)
• Formal definitions, worked examples, checkpoints, principles, war stories and reference workloads, indexed from the chapter headings
• A formulas-and-constants cheatsheet
• Generated glossary.md and patterns.md catalogs
• FastMCP tools, resources and prompts for BM25-ranked search and retrieval

Live counts are available from the ml-systems://index resource.

Quick start

Requires Python 3.12 or newer.

git clone https://github.com/CollinsNyatundo/machine-learning-systems-plugin.git
cd machine-learning-systems-plugin
python -m venv .venv
.venv/bin/pip install -r requirements.txt      # Windows: .venv\Scripts\pip
.venv/bin/fastmcp run mcp_server.py:mcp        # stdio; add --transport http for port 8000

To register the server with your AI clients in one step:

python scripts/install.py --dry-run    # preview
python scripts/install.py              # Claude Code, Claude Desktop, Cursor, Gemini CLI, Antigravity (those detected)

The installer backs up each config it edits and changes only its own ml-systems entry. See INTEGRATIONS.md for manual configuration and security notes. Set ML_SYSTEMS_DIR to use a skills/machine-learning-systems directory elsewhere.

MCP surface

Type	Name	Purpose
Tool	search_chapters	BM25-ranked search; compact snippets with file and heading
Tool	get_definition	Formal definition by term (exact, partial, then ranked)
Tool	get_artifact	Napkin math, checkpoints, principles, etc. by number or title
Tool	list_artifacts	Paged artifact titles, filterable by type and volume
Tool	get_chapter	Chapter text in pages, or one section; first page includes an outline
Tool	get_formula	Formulas and constants from the cheatsheet
Tool	get_cheatsheet_section	A cheatsheet section by heading, or the list of headings
Resource	ml-systems://index	Chapter metadata and live counts
Resource	ml-systems://glossary	Generated glossary
Resource	ml-systems://patterns	Compact artifact catalog (titles)
Resource	ml-systems://cheatsheet	Quick reference
Resource	ml-systems://skill	Plugin manifest (SKILL.md)
Prompt	study_chapter, compare_volumes, design_review, exam_prep	Structured study and review prompts

Tools return compact structured results by default and accept verbose (or paging arguments) for more. Every tool is read-only.

Content policy and maintenance

Chapters are the single source of truth. glossary.md and patterns.md are generated.

• Math uses \( ... \) (inline) and \[ ... \] (display) only. Bare $ and $$ are not allowed; literal dollar amounts are written \$.
• After editing chapters: python scripts/fix_content.py (normalizes delimiters and strips extraction artifacts), then python scripts/build_indexes.py, then python scripts/validate_content.py.

Known gaps in the extracted text: many Volume 1 and Volume 2 passages still contain raw Unicode math glyphs (for example 𝜏 opt) instead of LaTeX; Volume 2's own table of contents has no distinct Chapter 3 (numbering runs 1, 2, 4, 5 — a stray ToC row duplicates Chapter 4's title with no body of its own); chapter files begin with an extracted table of contents, and 21 of them also carry a verbatim "Full Slice Content" extraction after the condensed text, so some topics appear twice in search results. The original PDFs, when you have them, belong in docs/books/ (ignored by git).

Verify

pip install -r requirements-dev.txt
pytest -q
python scripts/validate_content.py
python scripts/build_indexes.py --check

Provenance

The chapter Markdown was extracted and reorganized from the CC BY-NC-SA 4.0 source material. Repository-specific changes include the directory layout, indexes, search server, metadata, and tests. Corrections to the textbook itself should be reported to the upstream project.

License

The source and this derivative are licensed under CC BY-NC-SA 4.0. You must provide attribution, restrict use to non-commercial purposes, and distribute adaptations under the same license.