Kavach

AI guardrails catch harmful prompts in English. Type them in Hinglish and some get through.

Kavach is a small shield that sits in front of any guardrail. It rewrites code-mixed input into plain English first, so the guardrail sees what is really being asked. Below: our pilot run, and the shield running live.

Pilot results

Loading results…

Try the shield

Type any Hinglish sentence. The shield only rewrites it into English. It never answers it.

Or try one:

Every prompt, every verdict

Harmful prompt text is never shown here, only its category and the guardrail's verdict. Safe prompts are shown in full.