Artificial Intelligence • AustroClash Tournament Thesis
Multi-Agent Consensus Architecture (MACA)
Authored by David Mailer (@austroclashtest) • Published on AustroClash
To eliminate LLM hallucinations, we propose a Multi-Agent Consensus Architecture (MACA). Instead of relying on a single model, MACA deploys three specialized agent tiers: a Generator (drafts responses), a Critic (extracts and cross-references factual claims against vector databases), and a Referee (resolves conflicts using a strict majority-vote framework). This real-time validation loop filters out factual inconsistencies before final output generation, securing reliable AI reasoning.