Every question written twice, so nothing is lost in translation.

The product

DriveTrain is a study app for the Russian driving-theory exam. It carries 800 questions organized as 40 papers of 20 — the ticket structure Russian theory practice is built around — and every question exists twice: text, all 2,504 answer options, and a written explanation, in both Russian and English. 548 of the questions come with a photograph or diagram of a road situation. All of it is installed with the app and works with no connection.

The problem I was solving

The source material for Russian driving theory is Russian, including its legal vocabulary, and its explanations cite paragraph numbers of the traffic regulations. That leaves a learner with two bad options: study in Russian and understand less than they think, or study an English translation and arrive at the exam having never seen the phrasing they will be marked on.

The obvious product answer — translate the app — solves the wrong half. Translating the interface leaves the questions in Russian. Translating the questions loses the terminology. DriveTrain holds both versions of the same question object and lets the learner move between them mid-question, with answers already given preserved.

Content and product decisions

• The parallel bank is the architecture, not a feature. Each question record carries both languages in the same object, so switching is a field swap rather than a fetch. That is why it is instant, and why it works offline and in every mode including the timed exam.
• Explanations are bilingual too, with no exceptions. All 800, in each language. A wrong answer that only tells you it was wrong teaches nothing, and an explanation a learner cannot read teaches nothing either.
• Answer options are translated individually. The distinctions between options are where theory questions are actually decided, so translating the question stem alone would have been cosmetic.
• Mastery is defined by the most recent attempt. A question counts as known only if you got it right last time, so a lucky guess later contradicted stops counting. The completion figure is therefore worth trusting.
• The mistake list maintains itself. Wrong answers collect automatically and leave the moment they are answered correctly — the highest-value revision list in the product, with no curation asked of the learner.
• Claims are deliberately constrained. The exam simulation is presented as DriveTrain’s own timed practice, not as a reproduction of the official test, and the app makes no affiliation or outcome claims.

What I contributed

Product strategy, the bilingual content model, the interface and its writing, and the build. The Android app is Kotlin and Jetpack Compose with a local relational database for questions, attempts, sessions, bookmarks, and exam results; the question bank ships as a bundled data file and loads on first run, so there is no content server and no account anywhere in the product.

Shipping the same bank on both stores meant the naming had to be settled in public rather than internally: the App Store listing reads “Drive-Train” and everything else — the app, Google Play, this site, the product site — reads “DriveTrain”. The site names both once so a searcher looking for either finds the same product.

Availability

On the App Store — listed there as “Drive-Train” — for iPhone, iPad, Mac with an M1 chip or later, and Apple Vision, requiring iOS 16 or later. On Google Play as “DriveTrain”, requiring Android 7.0 or later. Both are free to use, with subscriptions that unlock three additional practice modes. The consumer product site is at drivetrain.kanyantamubanga.com.