Misinformation Detector

Misinformation Detector is a concept-stage project: an agentic system that takes a claim, searches the live web for corroborating or contradicting sources, and returns a structured verdict with citations.

At a glance

• Problem: LLMs confidently state falsehoods. Fact-checking needs retrieval over current sources, not parametric memory.
• Approach: Agent loop — decompose claim → search → read sources → synthesise verdict with confidence and citations.
• Stack (planned): Claude API · web search API · structured output schema · citation grounding.
• Status: Concept — architecture sketched, not yet implemented as a portfolio demo.

Why this matters

The interesting engineering is not "ask Claude if it's true." It's orchestrating search queries, ranking source credibility, detecting when evidence is insufficient, and refusing to guess — the same grounded-generation discipline as DocSignal, applied to open-web retrieval instead of a fixed corpus.

What I'd build first

A minimal loop with a fixed claim set and hand-labelled expected outcomes — before any UI polish. Evaluation harness first, demo second.