LingoCall

Build your English. Prepare for your next interview.

English learning and interview preparation for people across industries.

Overview

A cross-platform mobile app that helps learners become fluent in spoken English by holding natural, real-time voice conversations with AI agents. Instead of flashcards or written drills, the learner picks a realistic scenario: chatting with a friend, calling customer care, sitting a job interview, or practising for IELTS, and simply talks. The AI listens, replies in a natural human-sounding voice, gently models correct grammar during the conversation, and produces a detailed report of how the learner did when the call ends.

The problem

Most language apps drill words and grammar in writing, but the skill that actually matters in interviews, customer calls, and daily life is spoken fluency. Learners practice for months and still freeze the moment a real conversation starts. There's no safe space to actually speak out loud with feedback.

How it works

Six conversation scenarios: Friend, Customer Care, Tech Interview, HR Interview, IELTS Speaking, Business, each driven by its own carefully-written system prompt and persona. ElevenLabs voice settings are tuned for lifelike delivery (low stability for expression, high similarity for consistency). A live reactive waveform on screen mirrors the learner's real microphone input. On call end, the model returns an explainable 0-100 score across Fluency, Grammar, Pronunciation, Confidence, and refuses to fabricate scores if the learner barely spoke.

What it does

  • Six conversation scenarios: Friend, Customer Care, Tech Interview, HR Interview, IELTS Speaking, Business, each with its own AI persona
  • Real-time voice conversation: STT → language model → TTS chained into a natural turn-taking loop
  • Live reactive waveform driven by the user's actual microphone amplitude via the Web Audio API: doubles as a mic diagnostic
  • Natural ElevenLabs voices tuned for human, expressive delivery; configurable accent (American / British / Australian / Indian) and gender + age
  • Passive grammar coaching: the agent never interrupts to correct, it reuses the correct form naturally so the learner absorbs it
  • Detailed post-call feedback: overall score 0-100, four sub-scores on a radar chart, mistakes with corrections, vocabulary upgrades, strengths and tips
  • Tech/HR interview verdict system: Strong Hire / Hire / Maybe / No Hire with per-question feedback
  • Progress + gamification: streaks, XP, levels, daily goals, achievements, weekly activity chart
  • Honest scoring: if the learner barely spoke, the app refuses to fabricate scores and shows a 'not enough to score yet' state
  • Demo Mode runs the entire experience client-side with no database: instant testing without API keys for everything except the AI calls

At a glance

Scenarios
6
Voice
ElevenLabs
Stack
React Native · Node · Postgres

Notes from the build

Built as a TypeScript monorepo (mobile / server / shared). The shared package holds the type contracts so client and server never drift apart. JWT auth (15-min access + 7-day refresh) with refresh tokens hashed in Redis. Passwords bcrypt-hashed at 12 rounds. Helmet, CORS allow-listing, Zod-validated inputs, expo-secure-store on device (Keychain / Keystore). Roadmap: native iOS / Android builds with on-device or Deepgram streaming STT, per-user usage limits + Stripe subscriptions, streaming TTS, saved call recordings.

Built with

React Native (Expo) + Expo Router · TypeScript · Zustand · Reanimated · react-native-svg · Node.js + Express · Pino · Zod · PostgreSQL + Prisma · Redis · Socket.io · OpenAI Chat Completions (gpt-4o-mini) · ElevenLabs (eleven_multilingual_v2) · OpenAI Whisper · Web Audio API (waveform)