Yarimai's Confluent Challenge

Project Name

YarimAI - Real-Time Viral Intelligence Platform

About the project

YarimAI is a content generation and sharing community powered by Google's latest AI models. Create stunning images and text with Gemini 3.0, generate videos with Veo 3.1, compose music with Google Lyria, and add professional voice narrations with ElevenLabs. Every creation becomes part of our vibrant community where inspiration flows freely.

Here's where it gets interesting: our remix functionality. See something you love? Click remix, and the original prompt and settings become your starting point. Modify it, regenerate, and create something new. The original creator automatically earns credits from your generation - building a chain of inspiration that rewards creativity across the community.

Here's where Confluent really shines in the user experience. On our generator page, when you select a content type - image, video, music, or voice - Confluent instantly serves the top 3 trending posts for that specific type. These aren't static recommendations; they're live viral content calculated in real-time. Click on any trending post, hit remix, and the original prompt becomes your starting point. This is Confluent enabling discovery and inspiration at the moment of creation.

In a content community, every interaction matters. But here's the challenge: how do you identify what's trending the moment it happens? How do you reward creators instantly when their work inspires others? How do you track inspiration chains in real-time? The answer: Confluent Cloud. Every user interaction - views, likes, comments, shares, and remixes - streams through Confluent Kafka topics. Our streaming service processes these events in real-time using Apache Flink, calculating trending scores, tracking engagement velocity, and updating rankings instantly. Built to scale from day one.

Architecture Diagram Labels:

User Interactions → Confluent Kafka Topics:
  ├─ user-interactions
  ├─ content-metadata
  ├─ view-events
  ├─ remix-events
  ├─ trending-scores
  └─ recommendations

Here's where Confluent powers everything. Our Kafka topics stream every interaction in real-time to this Viral Intelligence Dashboard. Watch the live updates indicator - it's active. We're tracking 138,000 total views, 916 interactions, and 44 viral posts. Content type distribution shows what's popular: images lead at 34%, followed by text and video. The Trending Now section displays 20 posts updated in real-time. Our Top 3 trending content shows live scores calculated by Flink. Popular creators are ranked by viral score - all metrics flowing through Confluent, processed instantly, displayed live.

Here's how posts become trending. When content is shared, a metadata event streams to Confluent. As users view it, view events flow through the view-events topic. Likes and comments add to the user-interactions topic. When someone remixes it, the remix-events topic captures the inspiration chain. Flink processes all these streams simultaneously, calculating a trending score using our time-decay algorithm: base engagement score multiplied by time decay, plus engagement velocity multiplied by five, plus a recency bonus. High velocity recent content gets boosted. The score updates in real-time, and when it crosses the threshold, the post appears in trending.

User interactions aren't just for ranking - they power our credit economy. When you remix someone's creation, Confluent's remix-events topic captures the entire inspiration chain. Flink calculates credit distribution across 10 levels: the original creator earns 10% of your generation cost, their inspiration earns 1%, and so on. All calculated in real-time. Creators see earnings update instantly as their work inspires others. This is Confluent enabling a fair, transparent, real-time creator economy.

YarimAI demonstrates Confluent Cloud powering real-time AI applications. Viral intelligence that surfaces trending content instantly, rewards creators fairly, and inspires the next generation. This is data in motion.