01 · Design
Prompts and test cases
Builds domain-specific scenarios in economics, finance, mathematics, and actuarial-style risk, including edge cases and multiple parameterizations.
AI training and model evaluation
Models do not earn trust. Evidence does.
As a Handshake AI domain expert, Junior develops and evaluates high-difficulty model tasks where economic logic, mathematical consistency, actuarial judgment, and clear reasoning must survive scrutiny.
Evaluation discipline
From prompt design to failure analysis
The work is not generic content review. It requires creating testable tasks, defining what correct reasoning must contain, locating failure modes, and producing reproducible evidence that can improve model behavior.
02 · Measure
Scoring rubrics
Defines correctness, required assumptions, methodological soundness, constraint compliance, and the evidence needed for a defensible score.
03 · Diagnose
Error taxonomies
Tracks hallucinations, unit and scale errors, base-rate neglect, causal confusion, bad constraints, and weak risk or valuation logic.
04 · Verify
Quantitative QA
Checks consistency, probability logic, expected values, discounting, regression interpretation, statistical inference, and sensitivity to assumptions.
05 · Communicate
Audit-ready feedback
Documents findings with reproducible reasoning, clear decision rules, and stakeholder-ready explanations while protecting confidential details.
06 · Generalize
Multimodal judgment
Applies structured evaluation across text, visual, and audio tasks, including temporal alignment, instruction following, language quality, and evidence use.
Technical coverage
Domains where small errors become large decisions
The differentiator is the combination of quantitative subject knowledge and evaluation discipline.
- Economics: macro indicators, causal claims, forecasting limits, and policy reasoning.
- Finance: time value of money, discounting, valuation, risk/return, and no-arbitrage intuition.
- Actuarial risk: expected loss, frequency/severity logic, pricing assumptions, tail risk, and model limitations.
- Mathematics and statistics: probability, scale, units, regressions, inference, sensitivity, and internal consistency.
- Communication: rubric-based writing, concise feedback, traceable reasoning, and professional tone.
Confidentiality boundaryThis page describes transferable evaluation methods and domain skills only. It does not disclose private datasets, proprietary prompts, unpublished rubrics, client information, compensation, or confidential project instructions.