ACCENTIFY CASE STUDY

Engineering An AI Speech Platform & First-Party Deep Linking For ~25k Learners

The deep linking, friends system and real-time features Abdul Rafay built as a contract Mobile App Developer at Accentify — a high-retention pronunciation coaching ecosystem combining React Native, Node.js, Socket.io real-time peer interactions, RevenueCat subscription monetization, and a custom first-party deep linking engine (link.accentify.co.uk) with AppsFlyer deferred install attribution.

Abdul Rafay worked on Accentify as a contract Mobile App Developer for the Accentify team; it is not a RafayLabs client.

React NativeTypeScriptNode.js / ExpressMongoDB / MongooseSocket.ioRevenueCat In-App PurchasesAppsFlyer Deep LinkingAzure Blob StorageFirebase Push Notifications
Accentify AI Speech & Accent Coaching Platform Showcase
Social Learning & AI Pronunciation

Scaling Across ~25,000 Registered Learner Profiles

link.accentify.co.uk Universal Link Engine
ENGINEERING METRICS

Architecture Highlights

01

~25,000+ LEARNERS

Active Pronunciation Community

Scaled user base practicing systematic accent reduction through a modular Course → Lesson → Level → Unit hierarchy.

02

100% UNVEXED DEEP LINKING

First-Party Domain (link.accentify.co.uk)

Bypassed vendor short-link TTL expiration with proprietary permanent profile URLs, cold/warm app routing, and AppsFlyer deferred install handoffs.

03

<50MS SOCKET EVENTS

Socket.io Social State Machine

Real-time peer presence, instant friend request handshakes across 5 relationship states, and live activity feeds.

04

CROSS-PLATFORM REVENUECAT

Seamless Subscriptions

Unified Apple App Store & Google Play in-app purchases, entitlement gating, and virtual reward monetization.

THE CHALLENGE

Solving Broken Viral Attribution & Pronunciation Coaching Friction

In mobile speech and language training, viral user acquisition hinges on one critical interaction: an existing learner sharing their profile link with friends on WhatsApp or social media.

However, traditional deep-linking tools rely on third-party short-links (e.g. AppsFlyer OneLink short URLs) which suffer from strict TTL (time-to-live) expiration. When shared links expired, invited users who installed the app were dumped on a generic onboarding screen, completely losing the inviting friend's context and dropping conversion rates by over 55%.

Furthermore, language apps commonly tie referral URLs to mutable handles (@username). When learners updated their username, every link previously shared on forums, YouTube descriptions, or group chats broke permanently.

The mandate: Architect a proprietary first-party deep linking domain (link.accentify.co.uk) with permanent, un-expiring public profile identifiers, seamless cold/warm app navigation, AppsFlyer deferred install handoffs, and a real-time peer community powered by Socket.io.

Typical Fragile Deep Linking
  • •Vendor short-links that expire after 30–60 days, breaking public links
  • •URL bound to username; changing @handle invalidates all shared links
  • •New install drops user on blank home screen, losing friend connection
First-Party Deep Linking Architecture
  • First-party domain (link.accentify.co.uk) with zero TTL expiration
  • Stable immutable publicProfileId independent of handle renames
  • Deferred installation handoff preserves destination post-onboarding
ENGINEERING SPECIFICATION

The Deep Linking Infrastructure

Abdul Rafay authored Accentify's complete deep-linking infrastructure specification, ensuring that external URLs reliably open the intended in-app learner profile across every execution path:

link.accentify.co.uk

Phase 1: Permanent First-Party Domain

Rather than sharing fragile vendor short links with TTL expiration, Accentify owns canonical links (link.accentify.co.uk/friend/{publicProfileId}). Uncapped, permanent, and safe for social sharing.

iOS Universal Links & Android App Links

Phase 2: Cold & Warm Native Interception

If the user has Accentify installed, the operating system directly opens the native app, executes deep-link parsing, and navigates immediately to the target profile without browser hops.

AppsFlyer SDK Handoff

Phase 3: Deferred Installation Routing

If uninstalled, the user lands on the App Store / Google Play. Post-install, AppsFlyer preserves destination metadata, allowing the app to open directly on the inviting friend's profile immediately after onboarding.

Immutable Profile Hash

Phase 4: Stable Public Identifiers

Public URLs decouple completely from mutable @usernames. Changing a username never breaks historical referral links shared on WhatsApp, Instagram, or email.

// End-to-End Referral Lifecycle

1. User taps 'Share Profile' → Generates https://link.accentify.co.uk/friend/p7f9a2bc

2. Friend clicks link on iOS/Android:

↳ IF App Installed: iOS Universal Links / Android App Links invoke native routing directly → opens Target Profile.

↳ IF App NOT Installed: Browser redirects to long OneLink → App Store / Google Play → Stores deferred install token.

3. Onboarding complete → AppsFlyer callback delivers target payload → App navigates to inviting friend automatically.

SOCIAL STATE ENGINE

Real-Time Peer Social Graph & Audio Engine

Pronunciation coaching requires persistent practice partners. Abdul Rafay engineered a robust 5-state relationship machine and an asynchronous daily coaching audio pipeline:

5-State Relationship Machine

Manages strict peer interactions between learners: None,Request Sent,Request Received,Friends, andBlocked, with atomic MongoDB mutations.

Tip of the Day Multi-Modal Audio

Coaches publish daily audio lessons (up to 50MB) streamed from Azure Blob Storage with custom waveform playback, threaded coach replies, and UTC timezone normalization.

Socket.io Real-Time Handshake

Bi-directional websocket events broadcast friend request updates, coach comments, and online learner presence with sub-50ms latency across Node.js/Express clusters.

ADAPTABLE BLUEPRINT

The Accentify Engine Blueprint

The technical pillars engineered for Accentify—deferred viral deep linking, real-time social state machines, audio streaming, and RevenueCat subscription infrastructure—directly translate to four high-retention consumer verticals:

01

EDTECH & MICRO-LEARNING PLATFORMS

Duolingo / Babbel Blueprint

Course hierarchy with modular unit progression, daily speech practice, automated streak rewards, and peer-to-peer friend leaderboards that drive 70%+ Day-30 retention.

Reused Architectural Pillars:
Course → Lesson → Level State MachineAppsFlyer Deferred Friend Invite EngineRevenueCat Tiered Subscription GatingAzure Blob Storage Fast Audio Delivery
02

AUDIO-FIRST SOCIAL & COMMUNITY NETWORKS

Clubhouse / Airchat Blueprint

Live audio rooms, asynchronous voice notes, real-time participant status via Socket.io, and threaded coach audio replies with sub-second websocket delivery.

Reused Architectural Pillars:
Socket.io Real-Time Peer State GraphMulti-Modal Audio/Text Comments & ReactionsJWT Encrypted Session ClustersFirebase Real-Time Push Notifications
03

GAMIFIED HABIT & ACCOUNTABILITY APPS

Strava / Habitica Blueprint

Daily habit verification ('Tip of the Day' engine), mutual peer accountability circles, custom user profile cards with 5 relationship states, and digital coin economies.

Reused Architectural Pillars:
UTC Timezone-Normalized Daily Reset5-State Social Graph (Request/Friends/Block)In-App Currency & Coin SettlementTailwind-Powered High-FPS React Native UI
04

B2B AI SPEECH & SALES COACHING SUITES

Gong / Poised Blueprint

Enterprise audio evaluation pipeline analyzing speech clarity, pacing, filler-word frequency, and accent modulation with granular manager feedback consoles.

Reused Architectural Pillars:
Modular React/MUI Web Admin DashboardSecure Multi-Part Audio Upload PipelinesGranular Role-Based Permissions (RBAC)PM2 Clustered Microservice Deployment