Case Study: Keek — AI-Powered Social Media Transformation
How Hanzo integrated AI infrastructure into the Keek social platform — content recommendation, multimodal moderation, and creator tools processing 12M daily content items.
Personas Social partnered with Hanzo to transform Keek from a traditional social media platform into an AI-powered content discovery engine. The integration spans recommendation, moderation, and creator tools — all built on Hanzo AI infrastructure.
The Challenge
Keek's chronological feed was losing engagement to algorithmically-driven competitors. Users spent less time discovering new content and creators struggled to reach new audiences. Meanwhile, content moderation at scale — millions of daily video, image, and text uploads — required an automated approach that manual review could not sustain.
What We Did
Recommendation engine: Adapted our EARLE recommendation system (originally built for commerce in 2014) for social content discovery. The engine combines collaborative filtering (users who engage with similar content), content understanding (multimodal embeddings of video/image/text), and behavioral signals (watch time, completion rate, shares) to build personalized feeds.
Multimodal content moderation: Deployed Zen vision models to analyze uploaded content in real time. The system classifies content across safety categories (violence, nudity, harassment, misinformation) with human-in-the-loop review for edge cases. False positive rate under 2%.
Creator tools: AI-assisted content creation powered by Zen models — automatic caption generation, hashtag suggestions, optimal posting time recommendations, and thumbnail generation for video content.
Infrastructure: The entire AI pipeline runs on Hanzo Cloud through the LLM Gateway, processing 12M daily content items with sub-200ms inference latency.
Results
- 400% increase in content engagement
- 85% reduction in harmful content reach
- 3x improvement in feed relevance scores (user survey)
- 12M daily content items processed through AI pipeline
- 60% reduction in moderation costs
"Hanzo didn't just give us AI features — they gave us AI infrastructure. The recommendation engine transformed how our users discover content, and the moderation system gave us confidence to scale without sacrificing safety." — Darius Mirshahzadeh, CEO at Personas Social
Hanzo AI is Techstars '17.
Read more
Zen: 1T+ Parameter MoE Architecture
Hanzo's flagship LLM — a 1.04 trillion parameter Mixture of Experts model fused from top language models, with distributed training and on-chain model attestation.
BitDelta: Serving 14 Zen Models from One GPU Cluster
How BitDelta's 1-bit delta compression lets us serve 14 Zen model variants from shared GPU infrastructure — the math, the architecture, and the tradeoffs.
From Commerce AI to Frontier Models: Hanzo's Decade of Infrastructure
Twelve years of building AI infrastructure — from a real-time recommendation engine in 2014 to 94+ frontier models and a post-quantum blockchain in 2026. The technical story of how we got here.