NARADA
Adversarial Multi-Model Deliberation Engine
What It Is
Adversarial multi-model deliberation engine (Claude / GPT-4o / Gemini / Grok) with Coherence Trap detection and Polymarket integration. Stress-tests AI reasoning through opposed debate rather than fragile single-shot prompting.
Single-model reasoning collapses into confident-but-wrong consensus. NARADA forces models with different training priors to actively disagree, and surfaces the coherence-trap failure mode when they don't.
What I Built
- Multi-model debate framework with role-asymmetric prompting
- Coherence Trap detector wired into the debate loop
- Polymarket integration for real-world prediction-market grounding
Tech Stack
ClaudeGPT-4oGeminiGrokPython
In active development.