Interviewer: You are the lead ML Engineer at a major payment processor like Stripe or PayPal. Design a system that evaluates transactions in real time to determine if they are fraudulent.
What are the questions you would ask first before answering this system design question?
@MyntraSupport, Not only has my issue been ignored despite CCTV proof of package not delivered ,but your system just emailed me someone else's confidential refund report&bank details.Your support response: "we don't want to upset you"over a severe data breach is unacceptable.
@MyntraSupport marked my order as delivered when I never received it. I shared CCTV evidence, repeatedly followed up via calls and emails No meaningful response just a closed complaint and a claim that I confirmed delivery over a call I never received.
Is this customer service
@ZypherHQ The rate cap issue on high-tier models is getting ridiculous. Paying top dollar just to hit "usage limit reached" after 4 prompt iterations makes it useless for real dev work.
Finance is the one sector where AI is uniquely powerful and uniquely catastrophic.
If you are building with LLMs, this is what protects your architecture:
1. Hard-coded data grounding boundaries
2. Isolated execution environments
3. Strict audit logging for every generated output
4. Human-in-the-loop approval gates
5. Fallback logic that defaults to legacy systems on edge cases
What else are we adding to this stack?