VERITAS

The Human + AI Truth Engine

Fighting Misinformation with Transparency

By Tobyn Bunch

🖼 IMAGE PLACEHOLDER

Logo: shield + checkmark / owl

💡 The Problem

Misinformation is everywhere — and current solutions are flawed.

▸ False claims spread faster than the truth on social media

▸ Fact-checkers are too slow to keep up with viral content

▸ AI systems are black boxes that give answers without showing their work

▸ People don't know who or what to trust

🖼 IMAGE PLACEHOLDER

Split screen: false social feed vs. confused person

🏛️ Our Inspiration

Veritas is the Latin word for "truth."

Our mission: to uncover truth in a world drowning in misinformation.

We wanted to build a fact-checking system that:

▸ Combines the speed of AI with the judgment of humans

▸ Shows its work — no black boxes

▸ Finds real evidence from trustworthy sources

▸ Returns a transparent score anyone can understand

🖼 IMAGE PLACEHOLDER

Dictionary 'Veritas' entry / justice statue

🔍 What It Does

Veritas takes any claim and returns a truth score with evidence.

▸ Users enter a claim (e.g., "Vaccines cause autism")

▸ Veritas searches .edu and .gov sites for evidence

▸ It extracts the full content from those sources

▸ Claude AI analyzes the evidence

Returns a 0–100 credibility score with:

▸ ✅ Clear reasoning (why the score was given)

▸ 🏷️ Status badge (Verified, False, Disputed, Uncertain)

▸ 📚 Source citations (where evidence came from)

🖼 IMAGE PLACEHOLDER

Veritas interface: claim, score, reasoning, sources

🏗️ The 3-Agent Architecture

Three AI agents working together.

🔍 Searcher Agent

Port 5001

▸ Finds trustworthy sources

▸ Priority: .edu, .gov, research, fact-checking

▸ Tools: DuckDuckGo API, trust-scoring

▸ Output: Ranked URLs + snippets

📄 Extractor Agent

Port 5002

▸ Reads full content from sources

▸ Tools: BeautifulSoup, browser headers, retries

▸ Challenge: Bypasses 403s & rate limits

▸ Output: Clean, readable text

🧠 Judge Agent

Port 5003

▸ Analyzes evidence & returns a score

▸ Tools: Claude AI, structured prompting

▸ Output: Score 0-100, reasoning,

▸ status, sources

🔄 How It Works — The Flow

👤 User submits a claim

↓

🔍 Searcher finds .edu / .gov sources

↓

📄 Extractor reads full content

↓

🧠 Judge analyzes with Claude AI

↓

📊 Return: Score + Reasoning + Sources

🛠️ How We Built It

Technology Stack

🖼 IMAGE PLACEHOLDER

Tech stack diagram with logos

⚠️ Challenges We Faced

Our biggest obstacles — and how we solved them

🏆 Accomplishments We're Proud Of

▸ Built a working 3-agent system in a single hackathon

▸ Real .edu/.gov source finding with trust-scoring

▸ Anti-blocking extractor that bypasses 403 errors

▸ Claude AI integration with intelligent reasoning

▸ Transparent output: score, reasoning, and sources

▸ Modular design — each agent runs independently

▸ Full monitoring with Arize OTEL tracing

🖼 IMAGE PLACEHOLDER

Trophy / checkmark icons per achievement

📚 What We Learned

▸ AI alone isn't enough — human judgment + transparent evidence map show the trust in the websites

▸ Monitoring is critical — Arize OTEL revealed what agents were doing

▸ API version management matters — Claude models get deprecated

▸ Modular systems are easier to debug — test each agent independently

▸ Source quality > quantity — one .edu beats ten random blogs

▸ Anti-blocking is essential — sites expect browser-like behavior

🖼 IMAGE PLACEHOLDER

Lightbulb / book icons per lesson

🚀 What's Next for Veritas

Future improvements we're excited about

🖼 IMAGE PLACEHOLDER

Roadmap / timeline graphic

🦊 The Verdict

Veritas: AI for speed, people for judgment, transparency for trust.

We built a fact-checking system that:

▸ 🔍 Finds real evidence from trustworthy sources

▸ 🧠 Analyzes with intelligent AI reasoning

▸ 📊 Returns a clear, transparent score

▸ 🏷️ Shows its work with source citations

The result: a tool that fights misinformation with transparency.

🖼 IMAGE PLACEHOLDER

Veritas logo / hero shot

VERITAS

Thank You

Uncovering truth in a world drowning in misinformation 🦊