neuroditto A mirror for your mind.

A private, local-first learning companion. It notices what you actually do, and reflects it back until you recognise yourself in it.

You read. You watch. You build. It scatters.

Here is one person’s week, as an example. Their stated goal: becoming an AI engineer.

3D Gaussian Splatting for Real-Time Rendering
Backpropagation, step by step
raytracer/ (half built)
A highlight from The Feynman Lectures
“Explain backprop like I’m five”
How a ray tracer works (for a friend)
NeRF: Scenes as Neural Radiance Fields
Shadertoy
Neural networks, chapter 3
A highlight on attention
“Why does my ray tracer look noisy?”
Gradient descent, explained to a friend

It reads along, and keeps the ideas.

Each paper, video, note or chat leaves concepts behind. When an idea comes back, it grows.

Takes in arXiv, YouTube, Kindle, Readwise, Zotero, GitHub stars, RSS, browser bookmarks and history, Markdown and Obsidian notes, ChatGPT and Claude exports, and local code.

The ideas link up into a map of how you think.

A link forms where two things share an idea. The more time you spend, the bigger the node.

In this week, most hours went to 3D graphics, and 9 of the 11 longest sessions were spent explaining something.

No grades. Only possibilities.

It sets what you said beside what you did, and offers a guess, with its evidence.

possible plausible likely

Likely — explaining systems is where you come alive.

Evidence 9 of your 11 longest voluntary sessions this month were teaching-shaped.

This week’s test One lunch-and-learn.

Try it: your answer tests the mirror, not you.

Not right? Change it. You hold the pen.

We never grade you.

No scores on people. Not now, not in a later version.

The mirror is the one that gets graded.

Every possibility is a logged bet. If it turns out wrong, its record goes down. Not yours.

You hold the pen.

Anything it says about you is a draft until you correct it.

Remember by returning.
Understand by explaining.

A flashcard comes back a little later each time. A Feynman coach asks you to explain it simply.

One next step. That’s all.

Today’s session

10-minute Feynman drill: Backprop

Because you’ve asked this question fourteen different ways.

how does backprop work?
chain rule in neural nets?
why do gradients vanish?
backprop vs gradient descent?
what is a computational graph?
backprop by hand
why multiply the gradients?
what is ∂L/∂w?
backprop in PyTorch
why does my loss explode?
intuition for backprop
backprop through a ReLU
what does .backward() do?
backprop, explained simply

Compass finds your why in evidence, not introspection.
Chapters is a quarterly story of your mind, which you co-write.

It lives on your machine. Your mind stays yours.

The graph, the record and the mirror all run on your own computer. Run a local model and nothing leaves, or bring one API key. Your choice.

Rough week? It’s context, not character.

Meet the mind you already have.

Want to know when it opens?