nasserDev/OwlBrain
Multi-LLM AI debate platform — orchestrate structured debates between Claude, GPT, and Gemini with consensus scoring and synthesized verdicts.
What's novel
Multi-LLM AI debate platform — orchestrate structured debates between Claude, GPT, and Gemini with consensus scoring and synthesized verdicts.
Code Analysis
8 files read · 3 roundsA multi-LLM debate platform where five specialized AI agents (Strategist, Analyst, Risk Officer, Innovator, Devil's Advocate) debate business cases across multiple rounds, with a consensus judge scoring convergence and a synthesizer producing a final verdict.
Strengths
Genuinely sophisticated multi-agent orchestration with consensus scoring, sycophancy detection, quality gates, memory compression, and clean provider abstraction across three LLM vendors. The prompt engineering, structured response parsing, and debate state management show real depth of implementation.
Weaknesses
The config store is a 58KB monolith that could benefit from splitting, and the regex-based response parsing is fragile if LLMs deviate from exact section formats. The orchestrator uses non-null assertions on captured state variables, a minor type-safety concern.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
42
Contributors
1
Files
159
Active weeks
2
Repository
Language
TypeScript
Stars
1
Forks
0
License
NOASSERTION