Loading…
Loading…
Curriculum
Every piece of the Koydo Lingua engine that an educator touches — eight things, shared across solo tutors, traveling tutors, and schools. The same curriculum, the same AI tutor, the same gradebook. The wrapper changes by persona; the engine doesn't.
1
144 units per language. Vocabulary banks calibrated to CEFR can-do statements. Conversation scenarios scripted by native-speaker curriculum authors. Pronunciation scoring on every audio response. Music-lesson tracks with karaoke alignment for pop-culture immersion.
2
Drills vocabulary, runs basic conversation rounds, scores written grammar, gives pronunciation feedback. No extra fee on Pro Educator or any School seat. Handles the repetitive 80% of in-lesson micro-interactions so the human teacher focuses on the curated 20%.
3
When advanced learners hit nuanced C1/C2 conversation, complex grammar derivations, or exam-prep simulation, Premium AI routes the request to a stronger reasoning model (Claude Sonnet 4.6 / GPT-5 tier). Sold once via the App Store IAP listing — not a separate marketing pitch.
4
On iOS 26 / macOS 26, common-path inference runs locally. Student utterances stay on the device. Cloud LLM falls back only when AFM is unavailable. Disallow-prompt-training honored end-to-end.
5
Every student carries a persistent progress record across in-home, online, and classroom sessions. Parents get a tutor-branded share-link with their child's reading-aloud audio, error rate, and mastered phrase count.
6
Snap a photo of handwritten work after a home visit. Lingua runs the tracing engine, OCR, and Whisper word-alignment if the student records audio. Everything attaches to the student's gradebook entry.
7
Built for the family living room where the WiFi is intermittent. Lessons cache, AI conversation runs on-device, parent share-links queue and sync when the tutor's iPad rejoins a network.
8
Lora display typography, warm-restrained palette, soft-shadow cards, gradient pill CTAs. Built to the Koydo Atelier canon. Designed to feel like a tool a working professional would choose, not a studio admin from 2014.
A language qualifies for the core 50 when we have native or near-native TTS, a curriculum author with C2-level fluency, and audio review by a second native speaker. Five preview-tier languages (Swahili, Telugu, Khmer, Nepali, Welsh) ship with a steered voice today and graduate when both the curriculum and the TTS pass review. For comparison: Lingoda Teams ships 14 languages; Babbel ships 14; Rosetta Stone Enterprise ships 24.
The 50-language curriculum covers the full CEFR scope (A1 through C1) in: Spanish, French, German, Italian, Portuguese, Mandarin, Cantonese, Japanese, Korean, Vietnamese, Thai, Indonesian, Tagalog, Malay, Arabic, Hebrew, Hindi, Urdu, Bengali, Tamil, Telugu, Punjabi, Gujarati, Marathi, Russian, Ukrainian, Polish, Czech, Slovak, Hungarian, Romanian, Bulgarian, Croatian, Serbian, Slovenian, Greek, Turkish, Dutch, Swedish, Norwegian, Danish, Finnish, Icelandic, Lithuanian, Latvian, Estonian, Albanian, Macedonian, Armenian, Georgian — plus US English for ESL learners. Five additional languages (Swahili, Telugu, Khmer, Nepali, Welsh) ship in a preview tier.
GPT-4o-mini runs in the background for vocabulary drills, basic conversation practice, grammar corrections on student writing, and pronunciation feedback on recorded student audio. It handles roughly 80% of the in-lesson micro-interactions a one-on-one tutor would otherwise type out by hand. The teacher remains the lesson planner, the curator, and the human relationship — the AI removes the repetitive grading and the late-night homework-review loop.
No. Premium AI is one in-app purchase option listed on the App Store IAP screen for tutors who need a more reasoning-capable model (Claude Sonnet 4.6 / GPT-5 tier) for advanced C1/C2 conversation, complex grammar derivations, or exam-prep simulation. It is part of the curriculum offering, not a separate marketing line.
On iOS 26 and macOS 26, the common-path AI inference runs locally via Apple Foundation Models. Student utterances never leave the device for the routine cases. Cloud LLM falls back only when AFM is unavailable or when the request specifically needs the larger remote model. Disallow-prompt-training is honored end-to-end.