Project Overview
Fanziz is a fast-growing sports and entertainment platform serving over 100,000+ content pieces across categories such as Cricket, Football, Tennis, Esports, Finance, Health, and Entertainment. With more than 1,000 new articles published daily, helping users discover relevant content became a significant challenge.
Yudiz designed and implemented an AI-powered recommendation engine that leverages semantic embeddings, vector search, and hybrid ranking techniques to automatically surface highly relevant articles in real time.
The result was a scalable recommendation platform capable of delivering personalized content recommendations in under 100 milliseconds while significantly improving user engagement.
Client Background
Fanziz is a fast-growing digital sports and entertainment platform featuring a massive library of over 100,000+ content pieces across verticals like Cricket, Football, Esports, Finance, and Health. Driven by a dynamic ecosystem, the platform publishes more than 1,000 new articles daily to engage its rapidly expanding audience. However, this massive influx of daily data created a critical challenge in content discovery, establishing an urgent need for an advanced system to seamlessly deliver relevant articles and optimize user engagement.
Business Needs
Our client wanted to create a highly scalable AI-powered recommendation platform that is secure, fast, and quicker than ever and weed out issues like information bottlenecks and slow content discovery. He wanted a platform that could enable real-time personalized recommendations aligned with the new trends of advanced semantic search and vector indexing. The idea behind crafting the platform was to invite his rapidly expanding audience to discover and engage with multi-category content aligned with the best performance, standards, and delivery conditions of the digital media market.
The Challenge
As Fanziz scaled its content library, the editorial team faced several challenges:
- Manual article recommendations required significant operational effort.
- Rule-based recommendation systems often surfaced irrelevant content.
- Traditional keyword matching failed to understand the context and meaning of articles.
- Users frequently consumed a single article and left the platform without exploring related content.
With over 1,000 new articles added every day, manual curation was no longer sustainable.
The goal was to build an intelligent recommendation system capable of automatically understanding article relationships and delivering relevant content recommendations at scale.
Our Solution
We engineered a semantic recommendation platform powered by OpenAI embeddings and Milvus Vector Search.
How It Works:
- Every article is processed through a background pipeline.
- OpenAI embedding models generate semantic vectors representing article meaning.
- Embeddings are stored and indexed using Milvus Vector Search.
- When a user opens an article, a hybrid search engine combines:Semantic similarity
Content relevance
Contextual ranking signals - The platform returns the most relevant recommendations in real time.
The entire recommendation workflow operates automatically without requiring editorial intervention.
Results
30% Increase in Session Duration
By presenting users with highly relevant content recommendations, overall session duration increased by 30%, indicating deeper engagement across the platform.
20% Recommendation Click-Through Rate
The recommendation engine achieved a 20% click-through rate, demonstrating strong relevance and user interest in suggested articles.
Scalable Recommendation Infrastructure
The platform successfully handles:
- 100,000+ content items
- 1,000+ new articles published daily
- Real-time recommendation delivery
Sub-100ms Response Times
Despite operating on a large content corpus, recommendations are delivered in less than 100 milliseconds, ensuring a seamless user experience.
Key Outcomes
- Eliminated manual recommendation workflows
- Improved recommendation quality through semantic understanding
- Increased content discoverability
- Boosted user engagement by 30%
- Achieved 20% recommendation CTR
- Delivered recommendations in under 100ms Built a scalable architecture capable of supporting continuous content growth
