quantsingularity/Agentic-AI-for-Anti-Money-Laundering-and-Regulatory-Compliance ? reverse-engineered prompt

Build me a working demo for an agent style AML compliance system that takes in transaction data, engineers useful risk signals, removes personal info before analysis, scores how suspicious each case is, checks names against sanctions and politically exposed person lists, then pulls the evidence together into a clean Suspicious Activity Report.

I want it to feel like a real analyst tool, not just a model script. Please include a simple web dashboard where an investigator can review the reasoning, see feature importance, read the generated SAR narrative, and approve or reject high risk cases with a human in the loop step. It should also support a streaming style demo, basic caching, experiment tracking, monitoring, and some adversarial tests that simulate evasion tricks like structuring and layering.

Use the sample or synthetic data in the repo so it runs end to end, preferably with Docker and a quick local demo option. If an API key is available, use an LLM for the narrative, otherwise fall back gracefully. Look up current docs online if you need to.