Application screening for bias
You receive nine hundred applications for every opening and recruiters skim each one for ninety seconds. Design an agent that does a structured first pass — extracting evidence against the role's actual requirements — knowing that a biased filter here creates legal exposure and shuts out good people. Tell me how you'd design it so it surfaces evidence rather than verdicts, how you'd test it for disparate impact before anyone relies on it, and then make the case to your executive team that this is defensible.
four_fifths_impact_flags(group_labels: list[str], advanced: list[bool]) → list[str][["group_a","group_a","group_a","group_a","group_b","group_b","group_b","group_b"],[true,true,true,true,true,false,false,false]]out["group_b=25"][["group_a","group_a","group_b","group_b"],[true,false,true,false]]out[][["group_a","group_a","group_a","group_a","group_a","group_b","group_b","group_b","group_b","group_b"],[true,true,true,true,true,true,true,true,true,false]]out[]Turn the fuzzy goal into a bounded agent task. Say what the agent reads, what it may do on its own versus draft for a human, and what it must never touch. Then earn the trust: how you’d verify it before rollout, and the pitch (the problem, what you built, why it matters).
Agent build: design the agent in plain language (or narrate it). The coach grades the decomposition, the guardrails, the verification plan, and the pitch.