Yassir Chat SDK

The SDK is structured as a clear boundary between a small public surface (ChatSDK entry point, ChatClient for runtime operations, and a ChatConfig.Builder for per-host configuration) and a much larger internal implementation. Each host app configures its own theming, feature toggles, and identity context, while sharing the same sync, networking, and persistence layers underneath — so improvements to the core benefit every integrating product at once.

• ● Builder-pattern configuration (ChatConfig.Builder) lets each host app opt into features and branding without touching internal code
• ● Dependency injection (Koin) isolates the SDK’s object graph from the host app’s, avoiding collisions across 6+ integrations
• ● Clean Architecture layering (data / domain / presentation) keeps the public API stable while internals evolve independently
• ● Versioned as an internal library so host apps can upgrade independently and roll back if needed

The SyncEngine is the core of the SDK’s reliability story. Every conversation has a local-first source of truth in Room, and a sync layer reconciles it with the server using different strategies depending on how far out of date the client is — from a lightweight incremental catch-up to a full resync after a long disconnect.

• ● Three sync strategies (none / incremental / full) chosen automatically based on a server-provided cursor, avoiding unnecessary full reloads
• ● A persistent outbound queue retries failed sends with exponential backoff (1s → 2s → 4s → 8s, capped at 30s), guaranteeing no message is silently dropped
• ● Background sync via WorkManager keeps conversations up to date even when the app is not in the foreground
• ● Optimistic local writes with reconciliation — the UI updates instantly and converges with the server state once sync completes

Real-time messaging runs over Phoenix channels on a single persistent WebSocket. I built the reconnection and authentication logic to handle the realities of mobile networks — tunnels, app switching, and expiring tokens — without ever leaving the SDK in an inconsistent state, even when multiple host app screens are using it at once.

• ● Exponential backoff reconnection (1s → 30s) with foreground/background awareness to avoid reconnect storms when the app is backgrounded
• ● Mutex-protected token refresh with double-checked locking — concurrent requests from multiple screens trigger only one refresh, with all callers sharing the result
• ● Atomic cache updates across the token manager and socket layer prevent races between an expiring token and a reconnect attempt
• ● Duplicate-connection guards and heartbeat monitoring keep exactly one healthy socket per session

The SDK extends beyond messaging into voice and video calls, using a hybrid architecture: call signaling (ringing, accept/decline, call state) travels over the existing Phoenix channel, while audio/video media is handled by LiveKit’s WebRTC stack. This reuses the SDK’s connection and auth layer for signaling instead of standing up a second real-time system.

• ● Call lifecycle (ring, connect, end, timeout) driven by Phoenix events, decoupled from the media transport
• ● LiveKit integration for WebRTC audio/video, with a 90-second timeout guarding unanswered calls
• ● Shared auth and reconnection logic with the chat layer — no duplicated token or socket management for calls