Best new learning, design and AI resources: day 29

Best Productivity Resources turns twenty-nine days old today. Yesterday's haul was a good one: a full self-study shelf for machine learning, some lovely design inspiration and tooling, and another wave of AI-agent and self-hosted utilities. Every resource below was reviewed by hand and links straight to its original website or repository.

1. Dive into Deep Learning, the interactive book

Dive into Deep Learning is an interactive open-source book that teaches deep learning with runnable code, math and discussion side by side. Use it as a course, a reference or a self-study path. Link: https://github.com/d2l-ai/d2l-en

2. micrograd, backprop in a few hundred lines

micrograd is Andrej Karpathy's tiny autograd engine. It's a few hundred lines that show exactly how backpropagation works under the hood, and it's one of the quickest honest ways to make neural networks feel less mysterious. Link: https://github.com/karpathy/micrograd

3. nanoGPT, small enough to read

nanoGPT is the simplest, fastest repository for training and fine-tuning medium-sized GPTs. The codebase is kept small and readable on purpose, which makes it a really good learning tool. Link: https://github.com/karpathy/nanoGPT

4. Neural Networks: Zero to Hero, Karpathy's video course

Zero to Hero is Karpathy's video course. It builds neural networks from scratch, one step at a time, and ends with a GPT. Pair it with micrograd and nanoGPT and you've got the full path. Link: https://github.com/karpathy/nn-zero-to-hero

5. Mathematics for Machine Learning, the maths you actually need

Mathematics for ML is the free companion book that covers the linear algebra, calculus and probability you actually need for machine learning. Read it next to a hands-on course. Link: https://github.com/mml-book/mml-book.github.io

6. The Stanford CS 229 Machine Learning cheatsheets

Afshine and Shervine Amidi's well-known CS 229 cheatsheets squeeze an entire Stanford machine-learning course into a handful of dense, printable references. Link: https://github.com/afshinea/stanford-cs-229-machine-learning

7. ML From Scratch, classic algorithms in plain Python

ML From Scratch builds classic machine-learning algorithms in plain Python with minimal dependencies, so you can read exactly how each one works. Link: https://github.com/eriklindernoren/ML-From-Scratch

8. Detail, for people who care about interfaces

Detail is a carefully made design resource for people who care about interfaces. The curation is sharp, and the presentation rewards slow browsing. Link: https://detail.design/

9. Toolcraft, for building your own design apps

Toolcraft is an open-source project from Pixel Point for building your own design apps with AI. It's a practical starting point if you want to ship a niche design tool. Link: https://github.com/pixel-point/toolcraft

10. Pryzm, a visual studio in your browser

Pryzm is a background and visual studio for designers that runs entirely in the browser. Handy for gradients, textures and quick visual experiments. Link: https://pryzm.design/

11. Valessa, 3D product shots without the pipeline

Valessa, built by Riotters, is a 3D product visualizer for making polished product shots and scenes without a full 3D pipeline. Link: https://valessa.riotters.com/

12. Book of Shapes, a gallery of generative patterns

Book of Shapes is a curated gallery of generative patterns. Calm, endlessly browsable, and a lovely source of visual ideas. Link: https://bookofshapes.com/

13. Iconly, an icon inspiration library

Iconly is an inspiration library of more than 2,400 icons from apps, games and web apps, organised so you can find a reference fast while you design interfaces. Link: https://iconly.design/

14. Stop Slop, for prose that reads human

Stop Slop is a skill file that teaches AI agents to strip the usual AI tells out of prose. Useful if you draft with a model and want the writing to read human. Link: https://github.com/hardikpandya/stop-slop

15. Speakr, self-hosted transcription

Speakr is a self-hosted audio transcription app that turns recordings into searchable text on your own hardware. Link: https://github.com/murtaza-nasir/speakr

16. AgentGlass, all your coding agents on one screen

AgentGlass puts every AI coding agent you run on one screen, with live cost and token tracking. Think of it as a practical observability layer for multi-agent work. Link: https://github.com/SirAllap/agentglass

17. Compass, a code knowledge graph in Rust

Compass is a local-first knowledge graph engine for code, written in Rust. It gives agents and developers a structural map of a codebase. Link: https://github.com/crabbuild/compass

18. Claude Deep Research Skill, research in eight phases

Claude Deep Research Skill packages an eight-phase, enterprise-grade research workflow as a Claude Code skill, so you can run it right inside your setup. Link: https://github.com/199-biotechnologies/claude-deep-research-skill

19. brag, a launch video from one command

brag is a command-line tool that turns a project into a shareable launch video with a single command. Nice to have on shipping days. Link: https://github.com/latent-spaces/brag

20. StateWave, for predictable app state

StateWave is an open-source memory runtime for AI agents that builds reproducible, provenance-tagged context bundles instead of relying on query-time retrieval, self-hosted on Postgres with pgvector. Link: https://github.com/smaramwbc/statewave

Also added yesterday, in brief

MatrixOne: a cloud-native database that brings transactional, analytical and AI workloads together. https://github.com/matrixorigin/matrixone

Claude Ops: operations tooling for Claude-centric workflows. https://github.com/Lifecycle-Innovations-Limited/claude-ops

slackcli: a command-line client for Slack power users. https://github.com/shaharia-lab/slackcli

The Art of Command Line: the classic one-page guide to getting really good at the command line. https://github.com/jlevy/the-art-of-command-line

Beam Weaver: Elixir-native agent tooling in the style of LangChain/LangGraph. https://github.com/caudena/beam_weaver

Build World: interactive 3D worlds with your AI agent. https://github.com/thrixel/build-world

Pretty Mermaid Skills: good-looking Mermaid diagrams as SVG or terminal ASCII. https://github.com/imxv/Pretty-mermaid-skills

gpty: a Godot-based Rust multi-PTY emulator. https://github.com/godot-pty/gpty

Ossium, Mimik, Raven, agtx, ModelScope Cookbook, Cybersecurity Projects, Swarm, Redamon, Portracker, Oya Browser, BrainAPI and vphone-ws: a fresh batch of self-hosted, systems and AI utilities, all searchable in the directory now.

That's the full set for day 29. Every addition is live in the directory with categories, best-for tags and pricing, and each one links straight to the source.