FameIntel

Transforming entertainment journalism from hours of manual research to 30 seconds of AI-powered intelligence.

Comment

What Inspired This Project

I kept seeing rampant misinformation online—users speculating about celebrity relationships without any verification. Twitter threads, Reddit discussions, TikTok comments—all based on rumors and hearsay. It struck me that there needed to be a way to actually verify these claims with real sources and confidence.

I also noticed something interesting: celebrity relationship data is highly interconnected. Taylor Swift connects to Travis Kelce, who connects to the NFL world, who connects to other celebrities—it's all a massive, evolving graph of relationships.

Perplexity was the unlock. With Sonar's real-time web access and trusted citations, I realized I could build the verification layer this space desperately needed.

How I Built It

I built FameIntel as a celebrity relationship intelligence platform using a graph database approach:

  • Memgraph to model 50,000+ celebrity relationships as an interconnected graph
  • open source embeddings on NVIDIA NIM for semantic search ("pop stars dating athletes" → Taylor Swift + Travis Kelce)
  • Perplexity Sonar Pro for real-time search augmentation and relationship verification
  • Perplexity Sonar Deep Research for comprehensive 15-minute intelligence reports
  • React + TypeScript frontend deployed on Cloudflare Workers for global edge performance

The architecture is simple but powerful: graph database for connections, vector search for discovery, and Perplexity for intelligent verification and research.

The Main Challenge

The biggest challenge was getting reliable markdown output from the Deep Research model. LLMs can be inconsistent with formatting, but I eventually cracked it using structured prompts and output validation.

I also needed to handle the long-running nature of Deep Research calls (15+ minutes). Using Cloudflare's Agents SDK, I built a monitoring system that gives users real-time progress updates instead of a blank loading screen. Users can see the AI working through comprehensive research in real-time.

What I Learned

  • Graph databases + vector search create powerful semantic discovery capabilities
  • Multi-model AI strategies work brilliantly thanks to Sonar Pro for speed, Deep Research for depth
  • Edge deployment matters for real-world performance (sub-second response times)
  • Structured outputs are essential for reliable AI applications

Impact Beyond Entertainment

While I built this for entertainment journalism, the implications go far beyond celebrity gossip:

  • Finance: Executive relationship mapping and network analysis
  • Politics: Political connection intelligence and influence tracking
  • Business Intelligence: Professional network analysis for any industry

FameIntel proves that specialized AI tools can deliver massive value by focusing deeply on one domain while building technology that scales horizontally.

The platform transforms 2+ hours of manual verification into 30 seconds of AI-powered intelligence, that's a 240x efficiency improvement with higher accuracy than manual research.

Built With

Share this project:

Updates