alihon20807-debug/forgeAI-hackathon ? reverse-engineered prompt

Build me ClaimGuard, a voice claims intake app for insurance calls that can handle code mixed Hindi and English conversation and safely turn a caller’s story into a claim workflow.

I want the app to take spoken input, redact sensitive stuff like Aadhaar, card numbers, and phone numbers before anything is sent to the model, then let the model suggest actions like opening a claim or staging a tow truck. The important part is that the system should not blindly follow the model, it should enforce safety rules so last minute cancellations, unauthorized waivers, and risky payments get blocked by the app itself.

Please include a simple supervisor style dashboard where I can see the live conversation, the current claim state, and a clear verdict on what was committed or held back. It should also log structured traces for evaluation and observability, with a clean setup that works locally and is easy to demo. If you need current docs or APIs, look them up online first.