Home with SRS Section
Scenario Header Collapsed
Scenario Dialogue Unlocked
Word Results
Sentence Results
Dialogue Turn 4 of 8 your turn
Dialogue Session Completed
SRS Card Hidden Recall from Memory
Progress Overview
Languages and Voice Sheet

Inspiration

Language learners often hit a frustration wall when transitioning from vocabulary apps to real conversational speaking. Traditional apps focus on text recognition, while hiring a private native coach for daily practice is expensive and inaccessible for most. We were inspired to build Vocally Live to bridge this gap: an app that delivers real-time, interactive spoken practice powered by localized AI models, giving users a private, immediate, and affordable way to gain real confidence speaking English.

What it does

Vocally Live is an AI-powered voice application designed to build true conversational confidence through structured, real-life scenarios.

Real-Time Voice Coaching: Users engage in natural, spoken conversations with an adaptive AI, receiving immediate feedback on pronunciation, phrasing, and speech clarity.

Contextual Scenarios: Practice realistic, everyday situations—from ordering coffee to professional job interviews—using structured, practical language frameworks.

On-Device AI Engine: By running localized Speech-to-Text, local LLMs, and Text-to-Speech directly on the device, user voice data stays private, fast, and accessible offline.

Frictionless Monetization via RevenueCat: Features a Vocally Pro tier that unlocks premium scenarios, advanced speech diagnostics, and offline voice packs using flexible monthly, annual, and trial subscriptions managed entirely through RevenueCat.

How we built it

We engineered Vocally Live with a focus on cross-platform performance, privacy, and seamless user experience:

Frontend: Built with Flutter & Dart for a responsive, native iOS and Android interface with real-time waveform visuals.

On-Device AI Pipeline: Integrated optimized local machine learning models (quantized LLMs, offline STT, and Kokoro/TurboQuant TTS) to execute directly on mobile hardware without cloud latency.

Backend & Infra: Powered by a lean Node.js / TypeScript microservices layer with Redis for fast session caching and state management.

Monetization Engine: Fully integrated the RevenueCat SDK (purchases_flutter) to handle entitlement management, dynamic paywalls, cross-platform subscription synchronization, and analytics out of the box.

Challenges we ran into

On-Device AI Latency: Balancing local speech synthesis and model response time on mobile hardware was tough. We had to optimize memory allocation and model quantization so that conversational turns felt natural and instant.

Synchronizing Real-Time Audio Flows: Managing low-latency audio capture, real-time transcription, and UI updates simultaneously without dropping frames required careful async state management in Flutter.

Paywall & Entitlement Architecture: Configuring multi-tier subscription access for offline vs. online models required a rock-solid entitlement architecture in RevenueCat to ensure accurate feature gating even when the device goes offline.

Accomplishments that we're proud of

True Offline Capability: Successfully running end-to-end voice processing locally on consumer mobile hardware while maintaining high conversational accuracy.

Smooth RevenueCat Setup: Deploying paywalls, trial offerings, and subscription entitlements in record time, allowing us to focus heavily on core product quality.

Natural Conversational Flow: Creating a fluid voice loop where users forget they are speaking to a local AI model and focus purely on practice.

What we learned

Local AI Execution: Deepened our understanding of mobile neural hardware acceleration, quantization strategies, and efficient audio buffer management.

Subscription Optimization: Learned how leveraging RevenueCat's dynamic paywalls and entitlement management dramatically simplifies mobile monetisation compared to building custom IAP validation servers.