I built an AI Neuro-Coach on FastAPI — here's what 60 cognitive trainers taught me about attention

Short summary

The author built WoW Brain, a neurocognitive training platform with 60+ adaptive trainers on FastAPI, and shares hard-won product lessons. Lab-grade adaptive algorithms fail in the wild; a hysteresis model that prioritizes user experience over statistical purity works better. Evening reflection loops boosted day-7 retention from 40% to 68%, and measuring anticipation (not streaks) predicted week-4 activity with 80% accuracy for high scorers.

• •Lab adaptive algorithms fail in production; hysteresis model with asymmetric difficulty adjustment works better
• •Evening reflection loop raised day-7 retention from 40% to 68% over morning-only check-in
• •Anticipation score (1-7) after third session predicts week-4 retention better than streaks or completion rates

Generated with AI, which can make mistakes.