Civic AI Safety System
AI that verifies
before it answers.
Verify AI uses a dual-agent pipeline — one AI generates, another audits — so community users only receive responses that have been verified for safety.
User Queryinput
GeneratorAgent 1
AuditorAgent 2
PASSsafe output
/BLOCKsuppressed
The Problem
AI sounds confident.
Even when it's wrong.
Standard AI systems generate one answer. There's no second check. In civic and community contexts — hospital details, government schemes, NGO contacts — a wrong answer can cause real harm.
Fake NGO Contacts
AI invents phone numbers and WhatsApp IDs for NGOs that may not exist.
Wrong Hospital Info
Incorrect addresses or policies for government hospitals that patients rely on.
Fabricated Schemes
Hallucinated government scheme names, portals, and eligibility criteria.
Overconfident Advice
Legal or medical guidance delivered as fact without real verification.
How It Works
Two agents. One truth.
01
User enters a civic question
No login required. Ask about hospitals, legal aid, government schemes, or community resources.
02
Agent 1 (Generator) creates a response
The generator agent uses Gemini to compose a helpful, informative answer focused on community guidance.
03
Agent 2 (Auditor) verifies the response
A separate auditor agent analyzes the output for hallucinations, fake contacts, and risky claims. Returns a structured PASS or BLOCK decision.
04
Safe output or warning is shown
If PASS: the verified response is displayed. If BLOCK: the response is suppressed and a safety warning is shown instead.