Broski AI Wingman

Broski AI Wingman is a mobile coaching app that helps users improve their messaging skills through AI-powered reply suggestions, screenshot analysis, daily challenges, and gamified progression. The client came with a frontend that had been built using AI code generation tools, a process sometimes called vibe coding, and the result was a codebase full of bugs, incomplete functionality, and no backend infrastructure at all.

The client had worked with Cenciss before. They returned for this project knowing what was needed: someone who could take a broken prototype, diagnose what was wrong, fix what could be saved, build everything that was missing, and ship the application to both the Apple App Store and Google Play Store.

Cenciss took full ownership of the frontend debugging, complete backend build, admin panel development, and cross-platform deployment.

The frontend codebase required triage before anything else. Navigation flows were broken, state management was inconsistent, and UI rendering errors appeared across multiple screens. The AI-generated code had structural problems that were not obvious from the surface, requiring a methodical audit of the entire codebase to identify what could be fixed and what needed to be rewritten.

Building the backend from zero was the largest single scope item. User authentication, AI service integration across multiple use cases, subscription state management, media handling, gamification logic, and content delivery all required a coherent backend architecture designed and built before any of it could connect to the front end.

OpenAI integration covered several distinct features: chat reply generation, screenshot analysis, rizz drill creation and scoring, confidence messages, and awkward conversation recovery. Each use case required its own prompt strategy, context management approach, and handling for token limits and response latency. Getting quality, consistent output across all of them required significant iteration.

RevenueCat subscription management on mobile is more complex than web billing. iOS and Android handle subscription state differently at the platform level. Webhook handling, subscription synchronization, trial period management, and cross-platform restoration all had to work correctly in both sandbox and production environments across both stores.

The image processing pipeline for screenshot analysis had to handle upload, temporary storage in Cloudinary, AI processing, and automated cleanup reliably. A broken cleanup workflow would accumulate storage costs. A broken upload would break one of the app's core features.

Gamification required a coherent system: XP calculation, level progression thresholds, daily challenge assignment, streak tracking, and rizz drill scoring all had to interact correctly without producing inconsistent or exploitable states.

App store submission for both Apple and Google added compliance and build configuration requirements on top of the development workload, with each platform having its own review criteria that the app had to satisfy.

The engagement started with a full frontend audit. Every broken navigation flow, state management issue, and rendering error was logged and prioritized before fixes began. The refactor was systematic: critical path bugs first, then secondary issues, preserving the original design intent wherever the underlying logic was sound.

The backend was built on Node.js and Express.js with a RESTful API architecture covering all application requirements. MongoDB schemas were designed for users, subscription records, challenges, videos, categories, and analytics data, with relationships structured to support the gamification and content delivery logic cleanly.

OpenAI ChatGPT Assistants API was integrated across all five coaching features, with distinct assistant configurations and prompt strategies per use case. Context management was handled to stay within token limits on longer conversation threads while maintaining response quality and consistency.

RevenueCat was integrated with webhook endpoints handling all subscription lifecycle events: purchase, renewal, trial expiry, and cancellation. Subscription state was synchronized to MongoDB and enforced server-side, ensuring feature access was always tied to the current subscription status regardless of platform.

Cloudinary managed the screenshot analysis pipeline: upload, AI processing trigger, and scheduled cleanup of temporary assets. The pipeline was built with failure handling at each step to avoid broken states.

The gamification system was engineered as a backend service: XP awards on feature interactions, level progression calculated against thresholds, daily challenge assignment on a schedule, streak tracking tied to daily activity, and drill scoring returned from the AI layer with persistence to the user record.

The React.js admin panel was built with JWT authentication, providing content management for videos and categories, user management, support request handling, and analytics dashboards for the client to monitor activity independently.

Expo build configuration was finalized for both platforms, with native module setup validated and compliance requirements met before submission to the Apple App Store and Google Play Store.