A social discovery platform powered by AI recommendation

Twistag built Joyraft, an AI-powered social event discovery app running on iOS, Android and web from a shared codebase, which reached 10,000-plus weekly active users with a recommendation engine that drives 40%-plus of clicks on events placed outside the top of the feed. Joyraft's problem was that good event matches went unseen when they fell below the fold. Twistag built React Native for mobile and Next.js for web behind a shared component library, and a hybrid recommendation engine on Google Cloud's Vertex AI combining collaborative filtering with content signals — then weighted friend activity above algorithmic confidence, because user testing showed social proof outperforms prediction.

Event discovery on mobile has a hard problem: users do not know what they are looking for, and recommendations that miss get ignored. Joyraft's founding team saw it directly in user research — when events were buried below the fold, even strong matches failed to convert because users never saw them. The discovery experience itself had become the bottleneck. The second constraint was structural. Joyraft needed to reach users across iOS, Android and web, and building three native applications meant tripling the work: design consistency broke down, feature releases staggered, and a small team could not ship fast enough to iterate on the AI layer at all. The business needed to grow and the platform could not scale with the team size.

Twistag built Joyraft as a hybrid stack — React Native for iOS and Android from a single codebase, Next.js for web, connected by a shared component library that enforced design consistency across all three surfaces. That architecture let a six-person team ship a feature once and see it live on three platforms simultaneously.

The recommendation engine sits at the centre. Twistag chose Google Cloud Platform's Vertex AI as the managed ML platform, so the team could iterate on the model without building infrastructure, and paired it with hybrid filtering: collaborative signals (what users like you engaged with) combined with content-based signals (event metadata, category, venue, stated interests). When a user marks interest or a friend attends something, the engine ingests that signal immediately.

The decision that shaped the product came out of user testing — social proof outperforms description. When users saw that friends had marked interest in an event, they engaged, so Twistag surfaced friend activity as the primary discovery mechanism rather than an afterthought, weighting it and public interest counts alongside algorithmic confidence instead of ranking on predicted rating alone. Twistag also solved cold start by sequencing rather than modelling: the app launched with no AI engine at all, behaviour accumulated for weeks, and recommendations were introduced gradually once the model had real data to learn from. The shared component library is what made three surfaces feel identical rather than merely similar.

Stack: React Native, Next.js, Vertex AI, Google Cloud Platform. Services: Product engineering, mobile engineering, AI integration and recommender systems, product design.

Joyraft reached 10,000-plus weekly active users across iOS, Android and web. The recommendation engine drives 40%-plus of clicks on recommended events that fall outside the top three positions in the feed — users engage with recommendations even when the algorithm places them low, which is the signal that the matching logic is trusted rather than merely prominent. Feature parity across the three platforms eliminated the release stagger that defeats most small multi-platform teams: one release cycle, one design system, three experiences that feel identical. The engine is trusted rather than merely prominent, which is the only durable form of discovery.

Forty percent of clicks land outside the top three — the engine works quietly.