A developer built an AI-powered French tutor to replace his $200/month human tutor, using spaced repetition and error tracking to improve learning retention at minimal cost.

What happened: The author created two linked tools — a knowledge-base tutor that tracks grammar topics using the SM-2 spaced-repetition algorithm (the same one Anki uses), and a voice app called Causons that practices conversational French by steering toward weak spots. The system logs specific error patterns (e.g., 'uses wrong auxiliary with motion verbs when sentence is negative — since 2026-04-12') rather than broad struggles, and adjusts review intervals based on a 1–4 recall rating after each topic.

Why it matters: Human tutors schedule by calendar; the author's software schedules by forgetting curve — reviewing topics right before they'd be forgotten rather than at random times. The author also found he was paying for a 20-minute conversation once a week when he could practice speech anytime. The system runs for a few cents per session by splitting providers: Groq (Whisper-large-v3 for speech-to-text at roughly $0.002 per minute), gpt-4o-mini for chat, and OpenAI tts-1 for text-to-speech, avoiding expensive APIs or per-token charges against a paid subscription.

What to watch: The author initially tracked every session separately but moved to a single JSON snapshot to manage file size, noting he may need to split it into several files in the future as the knowledge base grows. The voice pipeline processes three API calls in sequence (speech-to-text, chat, text-to-speech), with response times of 200–500ms per utterance on Groq's LPU hardware versus 800–1500ms on OpenAI's hosting of the same model.