Application · 2026 — Now

BitSize Languages

Japanese and French, from first sounds to real fluency.

  • RoleSolo · founder, engineer, designer
  • Timeframe2026 — Now
  • TypeApplication
14language apps across two tracks
N5–N1JLPT range for Japanese
A1–B2DELF range for French
Figure for BitSize Languages
Figure 1. Japanese and French, from first sounds to real fluency.

01The problem

Language apps reward streaks, not speaking — so learners can read but freeze when they have to talk.

02Approach & architecture

Fourteen apps across two tracks. Japanese: phonetics, listening, reading, vocabulary, kanji and grammar patterns. French: phonetics, listening, reading, verbs, from-English, difficulties, roots and grammar. Audio is generated with Edge TTS and checked by speech recognition before upload; listening and pronunciation lessons use in-browser speech recognition for speaking practice.

03Results

Measured What the work shows

Live at language.bitsizelearning.com with 3,100+ Japanese and 1,600+ French lessons, sharing one account and login with the rest of BitSize.

04Lessons

Browser-based speech recognition is good enough for pronunciation feedback — you don’t need a server-side STT pipeline for an MVP. JLPT levels as a content axis are stronger than free-form difficulty tagging because learners already know which level they’re studying for.

—Stack

DjangoPostgreSQLEdge TTSWeb Speech APICloudflare R2StripeRailway