Reading list of upstream repos I've cloned to study, dissect, or learn from. None of these are my work - they're listed publicly so the path I've walked is legible to people who care.
- NVIDIA/cutlass - CUDA Templates and Python DSLs for High-Performance Linear Algebra. The reference I keep going back to for understanding tile sizing and memory hierarchy on Hopper / Blackwell.
- siboehm/SGEMM_CUDA - Fast CUDA matrix multiplication from scratch. The blog post that goes with it is the cleanest "from naive to cuBLAS-class" walk-through I've read.
- KxSystems/kdb-taq - Latest source files for the kdb+taq package. Cross-referenced while learning q for tick-data work.
- karpathy/arxiv-sanity-lite - Tag arxiv papers of interest, get recommendations of similar papers using SVMs over tfidf vectors. My personal arxiv reader is downstream of this design.
- openai/parameter-golf - Train the smallest LM you can that fits in 16MB. Studied for the constraint engineering, not the model.
- Zjh-819/LLMDataHub - Quick guide for trending instruction-finetuning datasets.
- AmitSubhash/autolog - macOS ambient context capture that builds an activity knowledge graph from what you see, do, and switch between, then syncs it to Obsidian. Inspiration for OpsPilot.
- msitarzewski/agency-agents - A complete AI agency at your fingertips - frontend wizards, community ninjas, reality checkers. Useful taxonomy of role-shaped agents.
- Developer-Y/cs-video-courses - List of CS courses with video lectures. Curated my own self-study path from this.
- ibm-developer-skills-network/oaqjp-final-project-emb-ai - Watson NLP final project template from the IBM AI Engineering Coursera track.
- sstd521/Introduction-to-Linear-Algebra-5th-Edition---EE16A - Strang's textbook companion notebooks, used as a reference during EE16A-equivalent self-study.
Pure-clone forks of these repos used to live on my profile. They've been removed - starring an upstream is a more honest signal than carrying a stale fork.