While each project involves unique tasks, contributors may: Design and evaluate fee inquiry scenarios — verifying that fee amounts and category names match the bank's disclosed fee schedule exactly, and flagging cases where the system charge and the schedule diverge; Write statement clarification test cases that require distinguishing a descriptor-mapping question from a real fraud claim, and routing each correctly; Create autopay setup scenarios with deliberate traps: incorrect payee, insufficient-funds policy, wrong draft date, or missing confirmation steps; Build statement reprint cases that test realistic timing communication — including archived statements, fee citation, and the difference between self-service and manual retrieval channels; Develop beneficiary update scenarios that probe when a routine POD change crosses into ERISA/spousal-consent territory and requires specialist routing; Design card replacement cases with time-sensitive card-block urgency, fraud-signal recognition, interim mobile-wallet guidance, and delivery option trade-offs; Author account closure scenarios testing retention-offer eligibility logic, pending-transaction checks, and refusal to extend offers to ineligible customers; Write fraud claim warm-handoff cases — immediate card block, intake information capture, Regulation E clock awareness, and clean transfer to the disputes team; Grade responses on both factual accuracy and conversational tone: empathy under pressure, voice vs. Mindrift connects specialists with project-based AI opportunities for leading tech companies, focused on testing, evaluating, and improving AI systems.