Evaluate the AI's step-by-step process across calculations, ledger searches, and code to pinpoint quantitative or accounting errors. Create realistic financial tasks (e.g., valuations, backtests, audits) and write "gold-standard" examples to correct model near-misses. Build complex prompts featuring vague instructions, GAAP/non-GAAP differences, and conflicting estimates to test AI resilience. Create pass/fail checklists for deliverables and draft deceptive, polished-but-wrong answers to stress-test automated graders. Teach the AI to efficiently use core financial tools (SQL, etc) and categorize error patterns into reusable templates.
Job description provided by Google.
View the original posting
