VoiceClaim Auditor — Real-Time Claim Verification

💡 Inspiration

In fast-paced environments like startup pitches, academic lectures, and panel discussions, pausing to fact-check every claim is impossible. Misinformation or exaggerated metrics can easily slip by. We wanted to turn passive listening into an active, evidence-driven verification workflow.

⚙️ What it does

VoiceClaim Auditor is a real-time spoken claim verification system that listens alongside you. It seamlessly processes ongoing conversations, extracts verifiable facts, and researches them on the fly—all without interrupting the speaker.

Instead of oversimplified True/False labels, our system provides nuanced verdicts: Supported, Mostly Supported, Mixed, Misleading, Contradicted, or Insufficient Evidence. Every verdict is backed by an Evidence Confidence score based on source strength, contradiction levels, and data freshness.

The core UI is designed for continuous flow, offering live transcript updates, interactive claim cards, and real-time visibility into the research progress.

🏗️ How it works

The system is built on a sophisticated, multi-agent AI pipeline that breaks complex statements into atomic assertions. Here is how our verification engine works:

1. Transcription & Extraction: It captures and transcribes live speech using Speechmatics, instantly identifying objectively verifiable factual claims.
2. The Verification Pipeline: We use Bright Data to pull fresh web evidence, feeding it into a specialized trio of AI agents:
3. Evidence Agent: Gathers strong, relevant sources to support the claim.
4. Adversarial Skeptic: Acts as a red team, independently hunting for conflicting or newer evidence to challenge the claim.
5. Synthesizer: Weighs both sides objectively to produce the final nuanced verdict.

✨ Key Features & Modes

To adapt to different use cases, VoiceClaim Auditor includes customizable verification speeds and specialized domain modes:

• Verification Speeds: Choose Quick Verify for immediate feedback during rapid conversations, or Deep Verify for broader, adversarial web research.
• Investor Mode: Tailored for VC pitches to validate market sizes, growth metrics, revenue, customer claims, and competitive landscapes.
• Academic Mode: Checks research findings and benchmarks spoken claims directly against scientific literature.
• Custom Sources Mode: Allows users to upload proprietary PDFs, TXTs, or CSVs. The system will cross-reference live speech against these internal documents alongside live web research.

Built With

• education
• finance
• investment
• news-and-information
• recruitment
• scraping
• web