ThinkAloud

Learn the part AI can't fake for you

Build it yourself.Explain it out loud.

Pick a coding challenge and build it in your own editor and language. ThinkAloud shows you the concepts behind each step and the best places to learn them, then asks you to explain each one in your own words. The mentor helps whenever you ask.

Browse projectsSee how it works
  • Free, no sign-up
  • Speak or type
  • Never writes your code unasked

wc · explain it back

00:00

Prompt

Your wc has to count a 10 GB log on a laptop with 8 GB of memory. Why not load the whole file first, and how does your loop keep memory flat?

00:09

You · speaking

Loading it all would need 10 GB, more than the laptop has, so it would crash or start swapping. So I read a chunk, say 8 KB, into one buffer, count it, and reuse the same buffer for the next chunk. Memory stays about 8 KB whether the file is 1 MB or 10 GB.

00:41

Check · 3 points

  • Loading the whole file needs as much memory as the file
  • One buffer is reused for every chunk
  • Memory stays flat, whatever the file size
00:43

Mentor

read your answer → checked 3 of 3 points → looked ahead to milestone 3

Spot on. Here's what goes wrong with fixed chunks: é takes 2 bytes. What happens when a chunk ends between them?

Streaming in constant memory · explained
Projects fromCoding Challenges by John Crickett

Real tools to build, each linked to its original challenge.

codingchallenges.fyi ↗@johncrickett ↗

Why ThinkAloud

AI writes the code now. Understanding it is still your job.

Prompt, paste, run: it works until it doesn't, and then you're stuck with code you can't explain. The concepts underneath, and how your language really handles them, are what let you drive the AI instead of following it.

Prompt and paste

  • ×Eighty lines you didn't write and can't walk through
  • ×It runs, but you can't say why
  • ×A wrong answer looks exactly like a right one

With ThinkAloud

  • →You write every line, in your own editor and language
  • →You learn each concept from the best sources, then explain it in your own words
  • →The mentor tells you what's missing and where to read about it
Fundamentals aren't dead. They're your moat.

His post lists what's worth learning when AI writes the code: how networks really work, why databases lock, what breaks under heavy traffic, reading logs in a crisis, and being able to say you don't know yet.

Piyush Garg@piyushgarg_dev · 28 Oct 2026Read the post on X ↗

The part AI can't fake, and where you'll practise it

Each one maps to projects you can start today.

  1. 01How text really becomes bytes, emoji includedwcJSON Parser
  2. 02What memory is doing when the file is bigger than RAMwc
  3. 03How a request crosses the network, and what a status code promisesURL Shortener
  4. 04Saying "I'm not sure yet" without getting stuckJust say it in the chat

How it works

One project, four moves.

  1. 01 · about a minute

    Pick a project and your language

    Tell the mentor whether you build in Java, Python or Node.js. It looks up the classes and docs to study for each step.

  2. 02 · 30 to 60 minutes a step

    Build each step

    Each step says what to build, how to check it, and the concepts it uses, with links to the best places to learn them. Ask anything, any time.

  3. 03 · when a step works

    Explain it back, out loud

    The mentor asks you to explain what the step taught. Your own words are kept as proof you understood it.

  4. 04 · always yours

    Keep what you learned

    Export your learning log any time: every concept in your own words, the links you learned from, and the help you used.

The mentor

Help when you ask for it. Never before.

The mentor is an AI agent. It knows the project's steps and concepts, your progress, and what you've said, then decides how to help: a question, a nudge, a link to read, a small example. You choose how far it goes.

  1. Hinta nudge in words
  2. Similar casenot your code
  3. Talk it throughask anything
  4. Show me codeonly when you ask
least helpmost help

Just ask in the chat. Anything beyond a hint is noted in your help log, and you still explain the idea back yourself.

It will

  • Explain with words, analogies and small cases
  • Give whatever level of help you ask for, without making you prove you tried
  • Point you to the right page to learn from, not rewrite it

It won't

  • Write your solution unless you ask for it
  • Change your code. It only sees what you paste
  • Count a concept as explained unless your own words show it

With

Think out loud. Literally.

Explain by speaking instead of typing. Wispr Flow turns your voice into clean text in any app, including your answers here.

  • Explaining out loud is the quickest way to find a gap in what you know.
  • Speaking is faster than typing, so you explain the whole idea, not a short version.
  • Flow cleans up the pauses and restarts, so the mentor reads what you meant.
Get Wispr FlowReferral link: sign up through it for a 30-day free trial of Flow Pro. Flow also has a free plan.

Start with one

Real projects, not exercises.

Each one is a tool you'd recognise, built in steps of 30 to 60 minutes. Build it in Java, Python or Node.js: the concepts are the same in every language, and the mentor finds what to study in yours.

Challenge #1Not yet checked
stdin · any sizechunkchunkrestbuffer · 8 KBmemoryflat

wc

A command-line program that counts the bytes, lines, words and characters in a file, or in whatever is piped into it, and matches the Unix wc tool. It is small to write, but getting every count right forces you to understand how text is really stored and how a program reads input it can't hold all at once.

You'll learn text encoding, streaming in constant memory, counting words with a state machine

Challenge #2Not yet checked
{"ok": [1, 2]}{"ok":[1,2]}text in, tokens out

JSON Parser

A program that reads a file and says whether it is valid JSON, exiting with 0 if it is and 1 if it isn't, with a message that helps fix the problem. Building it step by step teaches the two stages every parser and compiler has, breaking text into tokens and then checking them against a grammar, and how a precise specification settles every edge case.

You'll learn grammar and parsing, strings, numbers

Challenge #12Not yet checked
long linkexample.com/blog/2026/how-dns-worksvisitor/x7Kqredirectshort code

URL Shortener

A small web service, like bit.ly, that turns a long URL into a short code, redirects anyone who opens the short link, and lets a link be deleted. It's an introduction to designing an HTTP API: choosing methods and status codes, generating short keys that don't clash, keeping data safe across restarts, and making requests safe to repeat.

You'll learn HTTP, the API contract, short keys

From Coding Challenges by John Crickett. 21 projects are ready, with more being written.

Questions

Before you start

Do I need to know a programming language?

Yes, the basics of one: Java, Python or Node.js. ThinkAloud is for going deeper, into the concepts underneath and how your language handles them.

Will it write my code?

Only if you ask, in short explained parts. It never changes your code.

Who checks my explanations?

The mentor, an AI agent. It marks a concept explained only when your own words show you understood it, and the sentence that shows it is kept as evidence. The app checks that sentence really is yours.

Do I need Wispr Flow?

No. Everything works with typing. Flow makes long explanations quicker to give, and it has a free plan.

What do I end up with?

A learning log: every concept you've explained, in your own words, with the links you learned from and the help you used. Export it as Markdown and keep it next to your project.

Is it free?

Yes, and there's no sign-up. Your progress is kept for your browser.

Pick a project. Start explaining.

Browse projects