VoiceClaim Auditor — Real-Time Claim Verification

VoiceClaim Auditor is a real-time spoken claim verification system built for situations where people cannot pause a panel, pitch, lecture, or discussion to fact-check every statement being made. It listens alongside the user, transcribes speech with Speechmatics, identifies objectively verifiable factual claims, and breaks complex statements into atomic assertions. Each claim is then researched against fresh web evidence using Bright Data, while AI agents evaluate both supporting and contradictory evidence. The verification pipeline uses an Evidence Agent to gather strong, relevant sources, an Adversarial Skeptic to independently search for conflicting or newer evidence, and a Synthesizer to compare both sides and produce a final verdict. Instead of oversimplified True/False labels, VoiceClaim Auditor returns Supported, Mostly Supported, Mixed, Misleading, Contradicted, or Insufficient Evidence. It also provides an Evidence Confidence score based on evidence strength, contradiction, and freshness. The core experience is designed for long, ongoing conversations. Claims are detected and verified progressively while the speaker continues talking, allowing users to see live transcript updates, claim cards, research progress, evidence, and verdicts without interrupting the conversation. Quick Verify prioritizes speed, while Deep Verify performs broader adversarial research. The MVP includes General, Investor, Academic, and Custom Sources modes. Investors can validate market, growth, revenue, customer, and competitive claims, while academics can check research findings and benchmark claims against scientific evidence. Custom Sources mode combines uploaded PDF, TXT, or CSV resources with live web research. VoiceClaim Auditor turns passive listening into an active, evidence-driven verification workflow.