Preply launches OpenAI-powered Lesson Insights to personalize language learning
Key Points
- AI-generated post-lesson summaries and feedback
- 95% weekly ChatGPT usage across Preply employees
- Personalized homework via integrated exercise engine
Summary
Preply uses the OpenAI API to generate AI-powered "Lesson Insights" that turn recorded 1:1 lessons into structured, personalized feedback and homework. With learner consent, sessions are transcribed, analyzed a few minutes before the lesson ends, and a report (grammar, vocabulary, pronunciation, summary, next steps) is delivered to the tutor and learner in-chat. Insights also feed a self-learning exercise engine to generate tailored practice that compounds across lessons.
Key Points
- Data flow: session recording -> transcription -> OpenAI API analysis -> insight report -> chat thread + exercise engine.
- Privacy: session recording and analysis occur only with learner consent; insights are returned within minutes after lesson end for real-time review.
- Integrations: OpenAI APIs power insight generation; outputs feed Preply Classroom, chat, and the personalized homework engine. Custom GPTs (e.g., Brand Voice) and Codex are used across the org for content consistency and engineering productivity.
- Operational notes: insight generation is scheduled near lesson end to ensure timely delivery; design emphasizes reliability and production readiness at global scale.
- Impact & metrics: 95% weekly ChatGPT usage among employees, ~75% of learners adopt Lesson Insights, >70% tutors use it, 75% retention of active learners after a year, 4.7/5 satisfaction from ~300k ratings, ~70% product-market-fit score.
- Rollout lessons: prioritize high-impact features, invest in enablement (workshops, prompts, datasets), partner with model providers, and embed AI as a cultural transformation rather than a one-off tool.
Engineering takeaways
- Build a low-latency pipeline for transcript ingestion, prompt orchestration, and structured output parsing to feed both UI and exercise-generation systems.
- Implement consent, auditing, and versioned prompt/config management for reproducibility and safety.
- Monitor usage and quality metrics (adoption, satisfaction, retention) to iterate on prompts and model configs; use Codex/AI assistants to accelerate engineering work on integration and tooling.
What's next
Preply plans to extend the learner model across months of study to provide continuous, individualized pacing, curriculum recommendations, and more precise feedback, enabling a human-led, AI-enabled learning experience.