The Inspiration A year ago, I started writing fanfiction with AI in Japanese. Not studying, just creating stories. Building narratives with characters I loved, crafting dialogue, describing scenes. Over time, something surprising happened: my Japanese improved naturally. Grammar patterns I'd struggled to memorize suddenly made sense because I'd written them into stories. Vocabulary stuck because words were tied to characters and emotions, not isolated flashcards.
That experience became the foundation for Animerse. What if we could give everyone that same natural learning loop, immersive storytelling where language acquisition happens as a side effect of having fun?
How Gemini Powers the Experience Animerse uses the Gemini 3 API as its core engine, delivering a multimodal Japanese tutor through an immersive scrolling story.
Structured Output drives every story beat. We use generateObject with strict Zod schemas to return tri-layer text (Japanese script, Romaji, English), tagged interesting words with formality annotations, and one of six interaction types, all as typed JSON. This ensures consistent, parseable learning content every time.
Audio Input enables pronunciation practice. Learners record themselves speaking Japanese phrases, and Gemini evaluates pronunciation accuracy. The feedback is woven naturally into the story, the NPC reacts to how well you spoke.
Vision Input powers handwriting evaluation. Learners draw Hiragana, Katakana, or Kanji on real paper, photograph it, and Gemini assesses stroke accuracy and recognizability, again, with feedback delivered through character reactions.
Long Context maintains story continuity. Each API call includes the full character profile, learner progress (vocabulary mastered, grammar seen, weak areas), and a rolling story summary, enabling adaptive difficulty without a separate ML pipeline.
Thinking is enabled for complex evaluation tasks to improve judgment accuracy on speech and handwriting assessment.
Gemini isn't a feature we bolted on, it IS the engine. Every interaction flows through it.
Log in or sign up for Devpost to join the conversation.