StanceStream – Real-Time Multi-Agent AI Debates Powered by Redis

This is a submission for the Redis AI Challenge: Beyond the Cache.

What I Built

StanceStream is a production-ready, multi-agent AI debate engine that transforms political discussions into live, evolving simulations.
Built entirely on Redis 8, it goes far beyond caching by orchestrating all four Redis data models in a single architecture to power intelligent agents with persistent personalities, memory-driven reasoning, real-time stance evolution, and advanced fact-checking.

The result is a fully operational real-time AI platform that demonstrates how Redis can serve as the primary data layer for a complex, high-performance AI application — handling everything from semantic search and live analytics to event streaming and structured data storage.

Demo

Live app: https://stancestream.vercel.app/
Repo: https://github.com/forbiddenlink/stancestream

Demo Highlights:
-4-mode navigation: Standard debate, Multi-debate viewer, Analytics dashboard, Business metrics
-Real-time stance charts tracking position changes during debates
-Semantic Cache Dashboard showing live Redis Vector operations
Multi-agent debates: SenatorBot vs ReformerBot on 8+ political topics
-Topic synchronization ensuring accurate stance tracking across all components

How I Used Redis 8

RedisJSON – Complex Data Structures
-Stores detailed agent profiles with nested personality traits, evolving political stances, and emotional states
-Tracks cache metrics, contest scoring, and AI-generated key moment summaries
-Maintains intelligence metrics for emotional trajectory analysis

Redis Streams – Real-Time Messaging
-Broadcasts public debate messages via WebSocket to all clients
-Maintains private agent memory streams for strategic reasoning and coalition building
-Logs message history with pagination for replay and context

RedisTimeSeries – Time-Based Analytics
-Tracks stance evolution over time for each agent
-Records emotional shifts influencing responses
-Captures performance metrics for live optimization and momentum tracking

Redis Vector – Semantic Intelligence
-Runs semantic caching with an 85% similarity threshold, achieving 70%+ hit rates
-Powers fact-checking by comparing claims across multiple knowledge bases using COSINE similarity
-Isolates topics to prevent cross-contamination between debates
-Uses OpenAI text-embedding-ada-002 for high-quality embeddings

Beyond the Cache
StanceStream demonstrates Redis as far more than a cache — it’s the real-time AI backbone for an application that merges natural language reasoning, fact verification, and live analytics.
By combining RedisJSON, Streams, TimeSeries, and Vector Search, the system delivers intelligent, personality-driven debates with business-ready dashboards, semantic intelligence, and millisecond-level responsiveness.

This project proves Redis can function as a complete, multi-model primary database for advanced AI workloads, pushing the limits of what’s possible in real-time applications.