| name | ml-library |
| description | Retrieve and cite from this machine-learning education corpus (923 docs — arXiv papers, course lectures, explainer articles). Use when the user asks an ML/deep-learning question, wants to learn a topic, asks for sources/papers/lectures on a concept, or wants a study path grounded in this repo. |
Retrieving from the ML library corpus
This repo holds 923 Markdown docs under corpus/ (papers/, youtube/, web/), each
with YAML frontmatter (title, url, tags, …). Your job: answer ML questions
grounded in these documents, with citations — never from memory alone.
Procedure
- Search broadly first. Full-text search is the most reliable signal:
grep -ril "<phrase>" corpus/ for concepts (e.g. "rotary position").- Narrow by tag when useful:
grep -rl "topic/<slug>" corpus/ (slugs in atlas/TAGS.md).
- For a learning path, open
atlas/topics/<Topic>.md (a curated Map of Content).
- Read the top 3–5 hits. Read each doc's frontmatter plus the relevant
sections (transcripts are long — don't always read the whole file).
- Synthesize and cite. Every claim → the source doc's frontmatter
url and
title. If sources disagree, say so. Never invent a citation.
- If tags miss, fall back to full-text search before saying the corpus lacks
something —
tags: are auto-assigned and approximate.
The 17 topics
neural-network-foundations classical-ml computer-vision
sequence-models-rnn transformers-attention language-models
efficient-architectures generative-models multimodal
reinforcement-learning alignment-rlhf reasoning-agents
efficiency-systems interpretability evaluation-trust
ml-engineering ai-industry-news
Rules
- This repo is a curation of others' work — attribute the original author, not
this repo.
- Prefer primary sources (papers) for technical claims; lectures/articles for
intuition and explanation.
- When the user wants depth, assemble multiple sources on the topic and trace each
point back to its origin (see
examples/self-attention-study-note.md for the
target output style).
