ConvoCompass
Language Learning Agent
THE PROBLEM
Language learning apps optimize for vocabulary and grammar drills, but conversational fluency requires something different: the ability to think and respond in real-time, in context. Most learners plateau because they practice recognition (reading/listening) but never production (speaking/writing) in authentic conversational settings.
THE APPROACH
Built a conversational agent that simulates real-world dialogue scenarios — ordering at a restaurant, negotiating a price, making small talk at a party. The agent adapts its complexity to the learner's level, provides inline corrections without breaking conversation flow, and uses voice synthesis for pronunciation practice. Session data feeds back into a learner profile that tracks progress across competency dimensions.
OUTCOMES
- Functional beta with voice-enabled conversation across 4 languages
- Adaptive difficulty system that adjusts mid-conversation
- Learner profiles tracking 6 competency dimensions over time
- Integration with ElevenLabs for natural-sounding voice interaction
KEY INSIGHTS
Interrupting a learner to fix a mistake breaks their flow and confidence. Inline, gentle corrections that keep the conversation moving produced better learning outcomes than precise grammatical explanations.
Structured scenarios with clear goals (order food, ask for directions) drove more engagement and measurable progress than open-ended 'talk about anything' sessions.
Adding voice synthesis transformed the product from a typing exercise into something that feels like genuine conversation practice. The medium is the pedagogy.