Hablas

The challenge

Most language apps are built to keep you tapping, not to make you fluent. I wanted to learn Spanish and then Portuguese the way the research suggests: material paced to the CEFR levels, plenty of comprehensible input, and spaced repetition tuned to how memory really decays, without a streak counter nudging me towards whatever keeps the app open.

What I built

• A Laravel, Inertia and Vue app with an FSRS-based spaced repetition engine. Spanish (Spain) and Portuguese (Portugal) decks are kept strictly apart and never mix in one review session.
• CEFR level tracked separately for reading, listening, speaking and writing. The headline level is the lowest of the four, not an average, so it cannot flatter me.
• A lesson engine with units, listening and speaking exercises, several exercise formats and a retake rule that sends you back to the weak spot instead of letting you scroll past it.
• Natural-sounding speech, generated ahead of time. Every spoken line is synthesised once with an open text-to-speech model, stored as a small audio file and served from the web server, with normal and slow playback and a choice of voices. Nothing is generated while you wait, and the audio sounds the same on every phone.
• Grammar mistakes tagged by kind (wrong gender, ser against estar, false friends, wrong tense) so review can target what actually goes wrong.
• An installable web app that works well on a phone, and a credits page for the voices and the licence of the model behind them.

Key decisions

• The weakest skill sets the level. A strong reader who cannot yet speak should not see an inflated badge.
• Pre-generated audio over live synthesis. It keeps the shared server light, makes every clip identical everywhere, and falls back to the browser voice where a clip is missing.
• Content is reviewed before it ships. New units stay hidden until a native speaker has checked them, and a test fails if Portuguese content drifts into Brazilian forms.