Personal SWE learning OS · Interview prep
Prepare for interviews
by building understanding
you can demonstrate.
Learn a mechanism, practise it in code or a diagram, explain it back in your own words, and let FSRS bring it back right before you would have forgotten it.
Free · no account needed · progress stays in this browser
Search & IR · concept
Tokenization
Explained back. Next review set by FSRS.
Illustrative · one concept moving through the loop
01 Learn
A mental model, a primary source, and the prerequisites.
02 Practise
Code or a diagram in a real workspace, run against tests.
03 Explain
Say it back in your own words. The AI asks; it never answers.
04 Review
FSRS brings it back right before you would have forgotten it.
Reading about a mechanism is not the same as being able to explain it.
That gap is the interview.
Every concept here ends in something you made or said: running code, a diagram, a decision, an explanation. Mastery is what you demonstrated, not pages you read.
One loop, every concept
Learn. Practise. Explain. Review.
DSA, system design, backend, AI systems, and behavioural rounds share one mastery model instead of five separate tools.
01 · Learn
Start from
the mechanism.
259 concepts across 19 tracks, each with a short mental model, a primary source, and the prerequisites it rests on. 26 roadmaps order them, from a 9-day reset to a 12-month run.
Concept
Tokenization
Splitting text into terms: lowercasing, stemming, stop words, n-grams.
- Mental model
- The same analyzer must run at index time and query time.
- Primary source
- Introduction to Information Retrieval, §2.2.1
- Related
- Inverted index · BM25
Illustrative · simplified from the app
02 · Practise
Make it run,
or draw it.
305 drills in a Monaco editor and an Excalidraw canvas. Code runs against tests; system designs become diagrams you can defend.
Build a search tokenizer
Lowercase, split on non-alphanumerics, drop stop words, stem. Same function at index and query time.
function tokenize(s) {
return s.toLowerCase()
.split(/[^a-z0-9]+/)
.filter(Boolean)
.filter(t => !STOPS.has(t))
.map(stem);
}Illustrative · simplified from the app
03 · Explain
Say it back,
without the key.
The Socratic companion only asks questions; it never hands over a solution. A Feynman explain-back gates the concept: one example, one failure mode, in your own words.
Socratic companion
If a document says “Foxes” and the query says “fox”, where in your pipeline do they become the same term?
Your explain-back
Both go through the same analyzer, so “foxes” stems to “fox” on both sides. The failure mode: stem too hard and “university” collides with “universe”.
One example · one failure mode
Illustrative · simplified from the app
04 · Review
Retrieve it
before it fades.
FSRS schedules the next retrieval from what you actually demonstrated, so strong concepts wait longer and shaky ones come back soon. 278 review questions keep it honest.
- TokenizationDue today
- Inverted indexIn 3 days
- BM25 scoringIn 9 days
Without looking: why must the tokenizer be identical at index and query time?
Illustrative · simplified from the app
What it covers
19 tracks. One mastery model.
From arrays and graphs to inference serving and agent systems, plus the interview round that is not about code.
Search & IR
Mathematics
Vector DB & ANN
AI Systems
Backend
Databases & Storage
System Design
DSA & Implementation
Behavioral & Communication
Go-to-Market
Systems Foundations
Infrastructure & Platforms
Distributed Systems
Inference & Serving
Agent Systems
Evaluation & AI Reliability
Developer Tools & Code Intelligence
Application Engineering
Multimodal & Spatial Computing
An honest fit
Interview prep that takes understanding seriously.
A fit
Engineers who want to explain why a B-tree, a rate limiter, or a KV cache behaves the way it does, and still remember it in three weeks.
Not a fit
An answer bank to memorise the night before. Nothing here hands you a solution, and progress is earned, not clicked.
A few honest answers.
Do I need an account?
No. Every route works as a guest, and guest progress stays in this browser. Google sign-in only adds progress that follows you across devices.
Does it cost anything?
No. There is no paid tier, subscription, or checkout.
Will the AI just give me the answer?
No, by design. The companion probes your understanding with questions. Working it out is the point.
Is it actively developed?
It is a mature personal learning system, built for its author’s own preparation and now in maintenance-only mode. It is kept working, not chasing features.
Start with one concept