Englify AI

THE PROBLEM

Most English learners practice in isolation — they speak, but nobody tells them what's actually wrong. Generic apps give scores without actionable insight. Non-native speakers plateau because they keep repeating the same mistakes without knowing it.

APPROACH

I combined speech-to-text transcription with LLM-based linguistic analysis to surface specific grammar and vocabulary patterns that a user gets wrong repeatedly. The product runs short speaking sessions, clusters errors by type, and serves targeted micro-exercises — not a generic curriculum, but a dynamic practice queue tuned to each user's actual weaknesses.

KEY DECISIONS

• 1

  Chose to build the error-clustering engine before the UI to validate that the signal was actually useful.

• 2

  Kept sessions under 3 minutes to maximize daily completion rates — longer sessions showed high drop-off in early testing.

• 3

  Used a freemium model: unlimited basic sessions, paywall on the personalized weakness reports.

• 4

  Shipped to a small beta group of Turkish expats first — the most motivated English learners with clear vocabulary gaps.

OUTCOME

Launched MVP in 2024. Early users report practicing 4–5× more per week compared to previous apps. The error-pattern model catches weaknesses that users were previously unaware of, with 78% of testers saying the feedback felt 'more honest than a human teacher.'