Show HN: Kaval – WhatsApp agent that checks if content is real or fake

I built a WhatsApp agent that verifies whether forwarded content is real or fake.

The problem: My parents and family keep getting increasingly sophisticated scam messages—fake traffic violation notices with convincing phishing links, AI-generated videos, "digital arrest" calls. They're not careless; the fakes are just good now.

How it works: Forward the suspicious message/image/video to the bot → get an instant breakdown of red flags (fake domains, AI-generated media, impersonation patterns, etc.)

Stack:

Meta WhatsApp Cloud API Claude for orchestration Gemini 3 Flash for content analysis Sightengine for AI image detection Modular architecture for swapping detection providers

Targets ~2-4 second response time for new content, sub-second for cached. The tricky part was balancing false positives—flagging legitimate content as fake would destroy trust faster than missing actual scams.

Try it for free: www.kaval.chat

Would love feedback, especially on detection edge cases. Happy to answer questions about the architecture.