#003Legend
Eigenfaces From Scratch: How Linear Algebra Learned to Tell 105 Celebrities Apart
10 min
Issue #005SeriesOct 20, 2026
> now reading: Issue #005
AI, explained in every shade.
Deep dives you can actually finish, projects with real code, and resources worth bookmarking. No hype, no fluff, just the good stuff, explained properly.
- Code-along posts with runnable notebooks
- 3 levels from Rookie to Legend
- Weekly new issues
- LLMs
- Agents
- RAG
- Diffusion
- Evals
- Fine-tuning
- Tokenizers
- Inference
- Computer vision
- MLOps
- AI safety
- Math for ML
> choose --level
Pick your level
Every post is tagged by difficulty, so you always know what you're getting into.
-
Rookie
New to AI. Start with intuition and pictures. No math degree needed.
Rookie posts -
Hero
You've run a notebook or two. Now learn how the internals actually work.
Hero posts -
Legend
Papers, math and production war stories. Bring coffee.
Legend posts
> ls ./blog --latest
Latest issues
#004Hero
Tokenization Explained !
10 min
#002Hero
Explaining Regularization L1, L2 and ElasticNet
12 min
#001Rookie
Ephemeral AI Agents : One job, then gone
TL;DR What are short lived or ephemeral AI agents?
16 min
No issues at this level yet. The next panel is still being drawn.
> git log --projects
Things we built
Real projects, real repos. Every one comes with a write-up of what worked and what didn't.
$ paperpal ask 2310.06825 "what is sliding window attention?"
→ answer + 3 citations (p.2, p.4, p.7)
Shipped
Paper Pal
Chat with any arXiv paper. Every answer cites the page it came from.
- Python
- FastAPI
- pgvector
$ python train.py --params 10M
step 4200 · loss 1.93 · "POW! said the robot"
Cooking
Tiny GPT
A 10M-parameter GPT trained from scratch on comic book dialogue.
- PyTorch
- Weights & Biases
$ arena vote --blind
model A vs model B · you picked B
Experimental
Eval Arena
Compare two models side by side and vote blind. Results feed a live leaderboard.
- Next.js
- SQLite
> open ./resources
Starter packs
Hand-picked papers, courses, tools and datasets, bundled by topic. Each pick comes with a note on why it's worth your time.
Weekly AI drops. Zero spam, all signal.
One email every Friday: the new issue, one project update and three links worth your time.