App for English Conversation Practice

This is a preview version and is still currently in development

I'm building a web application that helps German native speakers improve their conversational English through natural, spoken dialogue. The app adapts to individual learning needs with adjustable speaking speed, difficulty levels (A1-C1), topic selection, and instant mistake correction during conversation.

→ TRY IT HERE

Why this exists

Written English is no barrier today. German speakers translate and write fluently using AI and translation tools. But speaking is different. No existing AI application truly helps with conversational practice. I see this especially with people in their mid-50s (hello, dear parents!) who can read and write English but hesitate to speak.

Language learning apps like Duolingo offer some remedy, but you don't naturally talk in them, and they rarely give direct feedback during conversation. I've suggested people try ChatGPT's voice mode, but it speaks too quickly, uses overly advanced language, and jumps into complex topics without scaffolding. So I built something better.

What it does

The app creates a safe, psychologically supportive space for spoken English practice. Users select a topic (daily routines, travel, hobbies, work, etc.), choose their difficulty level, adjust speaking speed, and start talking. The AI responds with voice and text, following strict behavioral rules I designed based on second language acquisition principles:

• Turn-based speaking: Each prompt contains only the question and a short instruction ("Please answer" or "Please answer in one or two sentences"). No corrections, vocabulary lists, or tangents during the speaking turn.
• Gentle correction: After the user answers, the app gives short, warm feedback and gently corrects important mistakes with a simple explanation in English and a brief German clarification if needed. It then asks the user to repeat the corrected version.
• Adaptive difficulty: If the learner seems confident, the app uses slightly more complex sentences and richer vocabulary. If they struggle, it simplifies language, slows down, uses concrete examples, gives smaller tasks, and reassures them.
• Contextual vocabulary: The app tracks conversation context and periodically suggests 2-4 useful words or phrases connected to the current topic, with German meanings in brackets for difficult words.
• Natural conversation flow: Questions vary (open questions, yes/no followed by "why," compare two things, describe a situation, react to a statement) to keep practice engaging and realistic.

How I built it

I vibe-coded the app with Claude Code, combining my background in English Language and Literature with insights from working at PHBern in a didactic environment. The tech stack is Next.js with TypeScript, Tailwind CSS for a plain white minimalist UI, OpenAI's API for conversation and voice synthesis, and PostgreSQL for user progress tracking. The system prompt encodes strict linguistic and pedagogical rules: primarily English responses, German only when necessary, authentic encouragement ("That's a nice way to say it"), and psychologically safe framing (mistakes are normal and welcome).

What's next

This is a preview version. I'm working on user profiles to track learned vocabulary and recurring errors, improving the UI for better accessibility, adding pronunciation feedback, and building a progress dashboard that visualizes improvement over time. More features to come.