ryanzone/IntentGuard ? reverse-engineered prompt

Build me a Python app that acts like a safety layer in front of Razorpay orders for AI shopping agents.

I want it to take a user’s plain English purchase instructions, turn them into a signed policy, check the cart line by line, and then decide whether the order should go through, be blocked, or be put on hold for human approval. It should catch things like forbidden items, warranties, accessories, and anything that breaks the user’s budget or intent, not just the final total.

Please make it into a FastAPI backend with a simple dashboard for reviewing HOLD cases and showing the decision history. Use a Gemini based intent parser, HMAC signing for the policy, and keep an audit trail of every decision. It should be container friendly with Docker Compose, include sensible tests, and if you need current API details, look them up online.