Veritas

Veritas fights misinformation by combining AI speed with human judgment. It searches trustworthy sources, analyzes evidence with Agents, and returns a transparent credibility score.

Comment 1

Veritas is the Latin word for "truth." We chose this name because it represents our mission: to uncover truth in a world drowning in misinformation. Veritas takes any claim, searches .edu and .gov sources for evidence, extracts the content, and uses Claude AI to return a 0-100 credibility score with clear reasoning and source citations. It fights misinformation without trusting a single black box. We built Veritas using Python 3.12, TypeScript, and Flask with a main 3-agent architecture that functions as the backend code. A Searcher Agent (finds .edu/.gov sources), an Extractor Agent (reads full content), and a Judge Agent (analyzes with Claude AI). The agents communicate via REST APIs and run on separate ports. We encountered Claude model deprecation (404 errors), Arize OTEL setup complexity with BatchSpanProcessor, and 403 errors from websites blocking our Extractor Agent. Each challenge taught us something new. Our source trust-scoring system and anti-blocking extractor are features we're especially proud of. We learned that AI alone isn't enough; human judgment and transparent evidence are essential for trust. We also learned the importance of monitoring (Arize OTEL), API version management, and building modular systems that can be debugged and scaled independently. We plan to add a browser extension for one-click fact-checking, on-chain verification for transparency, custom evaluators, and support for multiple languages. Ultimately, we want Veritas to be the go-to tool for fighting misinformation everywhere.

Built With

Share this project:

Updates