Plenty of courses cover prompt engineering and RAG. This one is sharper where finance work actually fails: numerical validity, model disagreement, source discipline, bounded autonomy, and the evidence a controlled review requires.

✓ Two models, one contract Claude and GPT are compared on the same task, tools, data, output schema and budget.
✓ Verifier-backed Skills An independent recomputation runs beside the method, so speed never hides disagreement.
✓ Iteration between weeks Your Skil fails on an unseen case before the agent is built, not after the course ends.
✓ Evidence for review Tests, sources, limits and red-team results fal out of the build rather than a later scramble.

What you take away

Capstone & deliverables

You leave with exactly three artifacts you built for your own function:

✓ A dual-model guarded workspace and reproducible workflow
✓ A verifier-backed Skil with tests and an unseen-case report
✓ A bounded agent with evaluation and red-team evidence
✓ AIFI certificate of completion You also receive all twelve function-by-artifact reference implementations and the shared cohort Skill library.

Assessment & completion

Assessment is deliverable-based, not examined. Completion requires attendance at four of the five live sessions and submission of al three artifacts, including the final evaluation report.