Turning a photo or a recording into structured teaching feedback.

The problem

Candidates often prepare for Goethe, telc, ÖSD or TestDaF certifications alone. Classic apps mostly drill vocabulary, but offer little feedback on a handwritten essay or a spoken answer — precisely the exercises for which an outside eye is most useful.

The experience designed

Written — the user photographs their handwritten essay. The app identifies the content and returns a structured evaluation based on the level and the target exam’s rubric.

Spoken — the user records their answer. The system analyses the audio and returns a skill-by-skill evaluation with improvement paths.

Practice — exercises follow real exam formats, with timing, a sample bank and spaced repetition for vocabulary.

End-to-end contribution

Problem observation, value proposition definition, user-flow design, Flutter development, orchestration of the multimodal AI, cost model and credit system, Next.js admin, landing page, SEO content, analytics, store publishing and user support.

Key decisions

Evaluate rather than merely generate

The AI is not used to produce a generic lesson. It is guided by an evaluation rubric and an exam context so feedback stays comparable from one attempt to the next.

Bake cost into the product

Vision and audio processing has a variable cost. Freemium, credits and usage limits were designed alongside the experience so the product stays sustainable.

Measure real engagement

Dedicated instrumentation tracks session length and drill usage. This lets us tell a downloaded app apart from an app actually used to revise.

Results

Deutsch Exam has passed 3,000 users with an average session length of 1 h 03. The 4.8/5 rating and feedback tied to exam success confirm the perceived value.

What this project proves

• full autonomy, from concept to support;
• integrating multimodal AI into a useful flow;
• ability to ship in a month and iterate;
• understanding of the business model;
• Flutter, Firebase and acquisition web.