Nidhi2905/Ai-Security-RBAC-Sandbox ? reverse-engineered prompt

Build me a small local Python app that shows how AI access control should work for company data.

I want it to use a mock database file and a local Ollama model, but only after the app checks who the user is and what role they have. If someone is not allowed to see a record, the app should block it before the AI gets any sensitive data. Please also simulate a few prompt injection attempts so I can see the difference between a vulnerable setup and the protected one.

It should log every request to an audit file with a clear granted or denied result, plus a timestamp, so it feels like a simple security demo for IT and compliance. Keep the app easy to run locally, and if you need to look up current Ollama docs online, go ahead. I also want it to be clear which data is safe to show and which data must stay hidden.