How to run

Dependencies and the Python version are declared in pyproject.toml, with exact versions pinned in uv.lock. Library code lives in src/ai_learning/ (importable as ai_learning); notebooks live in notebooks/. Requires Python 3.12+.

We recommend uv, but it's not required — any standard venv + pip works too, since everything is in pyproject.toml.

With uv

uv reads the lockfile, so you get the exact same versions as everyone else.

uv sync                  # create .venv and install everything (incl. dev tools)
uv run jupyter lab       # run a notebook
uv run pytest            # run tests
uv run python -m ai_learning   # run package code

No need to activate the venv — uv run handles it (this also avoids the Windows-vs-Unix activate-script difference).

You can also open a notebook in VSCode if you have the extension, just make sure you pick the Python kernel from the venv.

With pip

Skip uv entirely if you prefer. You won't get the locked versions, but pyproject.toml still defines compatible ones.

python -m venv .venv
source .venv/bin/activate        # Windows: .venv\Scripts\activate
pip install -e .                 # editable install (runtime deps)
pip install --group dev          # optional: dev tools (needs pip >= 25.1)

Then run things directly (jupyter lab, pytest, etc.) with the venv active.

Guidelines

Main is for abstract interfaces (ex. Tokenizer, Data Samplers)
Personal repositories will be playgrounds/sandboxes

AI Resources

This is a repository which aims to collect resources on learning about AI, how to use it effectively, the ethics of using it, and the capabilities of free and open source models.

How AI works

This section is for resources pertaining to how AI works.

Books

Hugging Face LLM Playbook

• Hugging Face

Build A Large Language Model

• manning
• github

Videos

3blue1brown: Deep Learning

3blue1brown has a series on Deep Learning that's really good, the chapters below on LLMs/transformer don't rely too much on the preceding sections.

• Transformers, the tech behind LLMs | Deep Learning Chapter 5
• Attention in transformers, step-by-step | Deep Learning Chapter 6

Effective Use

Ethics of AI

Podcasts/Articles

The Most Important Question Nobody's Asking

• dwarkesh article and podcast

Free and Open Source Models

Nvidia Nemotron 3

• research report

Google Gemma

• gemma

Tools

Allium

The Allium CLI validates specs and catches structural issues such as missing transition witnesses and unreachable triggers. It also generates tests from specs.

• Github

Pi.dev

Pi is a minimal terminal coding harness. Adapt pi to your workflows, not the other way around, without having to fork and modify pi internals.

• Github
• Pi.dev