The ONLY Data Structures and Algorithms interview prep you need.
by Elman
Stop failing them. Months of preparation, then an unfamiliar problem, a minute of nothing.
No account, about seven minutes
A weather log lists up to 100000 daily temperatures. For each day, find how many days you must wait until a strictly warmer day appears (0 if none does).
Which pattern?
5s
No heading, no hints. Same as the interview.
REVIEWS
What learners say
- steps
- 150steps
- topic groups
- 19topic groups
- patterns
- 52patterns
- recognition prompts
- 194recognition prompts
- languages
- 6languages
- reading depths
- 3reading depths
WHY IT GOES WRONG
Algorithms aren't hated because they're hard. They're taught badly.
01 Every list told you the answer in the heading
The problems you drilled sit under the name of the pattern that solves them. You practised writing the code and never practised choosing it.
“I already know the category it belongs to, so my brain is biased towards using that specific pattern.”
02 So the interview spends its first 20 minutes on naming
Not on the code. On working out what kind of problem this is, while the clock runs and someone watches.
“When I open a new Graph question, I often spend 20-30 minutes just trying to figure out what kind of problem it is.”
TWO ROADMAPS
The usual route loops. This one finishes.
Both roads arrive at the same thing: a problem you have never seen. One of them prepared you for it.
How people usually prepare
No stored answer, and no way to work one out. Then the list again.
The AlgoPath path
The problem is new, the type is not. You name it and work the answer out.
WHERE YOU STAND
Fifteen problems, and the test tells you where to start.
About seven minutes, no account, no code. You read a problem and name the pattern it needs.
- The right pattern for every one of the fifteen
- Which groups held, and which came apart
- The step to pick the course up at
No card and no email. It marks nothing as done.
HOW ONE STEP GOES
How each topic works
AlgoPath is a pattern-recognition trainer for coding interviews, with a 150-step course under it.
Start from something real
A cache that keeps filling up, a log you have to scan once. Not a puzzle.
Get the idea in plain words
Then the same idea in a developer's terms, or formally. You pick.
Write the code yourself
In the browser, against the step's real tests. Reading someone else's answer does not count.
Meet it again with no label
The trainer takes the heading off and asks you to name it, weeks later.
Steps 00-08 assume nothing and are the way in. The first 33 are free in full.
PICK YOUR LANGUAGE
Write it in JavaScript, TypeScript, Python, PHP, Java or C#
Same exercise, same tests, six languages. Every one of them runs inside your browser, so nothing installs and no server ever sees your code.
JavaScript
The default. Your code runs in a Web Worker against Node-style assertions.
TypeScript
The same exercise with the types on. They come off in the page, so what runs is the JavaScript worker.
Python
Real CPython in the browser via Pyodide. Write it in idiomatic snake_case.
PHP
Actual PHP 8.3 compiled to WebAssembly. The same cases, zero setup.
Java
A real javac and a real JVM, through CheerpJ. It compiles in the page before it runs.
C#
The .NET runtime and Roslyn, both in the tab. Your code is compiled the way it would be on your machine.
AFTER TEN OF THOSE
You get a list of what you mix up with what.
194 prompts across 52 patterns. The table below is what a few sessions leave behind.
Ten in a row
Ten prompts, one clock, no verdict until the end.
Signals, not vibes
After each answer, the words in the statement that gave it away.
Weakest first
The top row is the next thing to practise. It links to its step.
| Pattern | Recognised | Confused with |
|---|---|---|
| Monotonic stack | 40% | Sliding window |
| Binary search on the answer | 62% | Greedy |
| Union-find | 75% | Graph BFS / DFS |
| Two pointers (opposite ends) | 95% | nothing yet |
PICK YOUR DEPTH
One lesson, three depths
A real paragraph from the course, in the three depths it was written at. Switch between them, and the choice sticks for the whole track.
You have a row of items and you want to rearrange them without a second row to work in. Think of tidying one shelf of books. You do not clear an empty shelf next to it. You slide the books along the shelf you already have, swapping two at a time.
You reorder the elements inside the array's own storage, with no second array to stage in. One index marks where the next kept element goes; another scans forward looking for elements to keep.
The transform operates on a single backing store of length n. A destination index and a scanning index traverse the array, and elements are relocated by in-place assignment or swap. Auxiliary space stays O(1) because no allocation scales with n.
WHERE IT RUNS
Every algorithm runs in software you already use
This is the difference from a puzzle site. Each pattern is wired into a small production case, so you see the system it belongs to.
Rate limiter
A sliding window keeps an API from being hammered.
LRU cache
A hash map and a linked list decide what stays in memory.
git bisect
Binary search finds the commit that broke the build.
Autocomplete
A trie turns each keystroke into suggestions.
WHO IT IS FOR
Everyone enters at their own speed
An interview in a few weeks
You have solved a few hundred problems and still stall on one with no heading. Take the test, start where it says, and spend the weeks on recognition rather than a fourth pass through arrays.
Starting from zero
No CS background, and every course you opened started in the middle. Steps 00-08 assume nothing and explain each term the first time it appears.
FAQ
Straight answers
What is AlgoPath, exactly?
AlgoPath is a pattern-recognition trainer for coding interviews, with a 150-step course under it. The trainer gives you problems with the label removed, times you, and records which patterns you swap for which. The course is where you go for the ones you could not name.
I have already solved 200 problems. What does this add?
The lists you solved were grouped by pattern, so the heading answered the question for you every time. Here it is gone, and you find out which patterns you actually confuse.
Is this another problem site?
No. A problem site is a bank of problems and a judge. This trains the minute before you start typing, and keeps a record of how that minute goes.
Why pay for this when an LLM will just solve it?
It will, and it will explain the solution well. Then the interview starts and the AI is switched off: 62% of companies still prohibit it (Karat, January 2026). Choosing the algorithm is the one skill an LLM takes off your hands while you practise.
I am starting from zero. Is this for me?
Yes, through the course rather than the trainer. Take the placement test and it will say which step to start at. If that is step 00, it explains what an algorithm is before anything else. The trainer starts paying off once you have patterns to confuse.
Is this for interviews or for real work?
Both. Every step names the production system its algorithm runs in, so it sticks past the interview and you stop relearning the same topics before every job hunt.
How much is free?
The placement test, with no account at all. Then the first 33 steps in full: lesson, runner, tests, progress. No card.
One payment, both halves.
The trainer and the course. No subscription, no renewal.
Join the free beta
Sign up now and get 60 days of full access, free.
Start with the one you cannot name.
Fifteen problems, about seven minutes, no account.
Take the placement test