WorkCase 04

Bayesian Hiring Agent

Cost-aware, fairer resume screening by orchestrating multiple LLMs under uncertainty.

  • agents
  • fairness

p. 01

The case

Context, problem, approach, result.

Context

Written up as a sole-authored preprint — "Bayesian Orchestration of Multi-LLM Agents for Cost-Aware Sequential Decision-Making", arXiv 2601.01522.

Problem

Screening at volume forces a choice nobody wants to make explicit: give every candidate the same expensive model pass, or give them all the same cheap one. The first does not scale and the second is where unfair outcomes hide.

Approach

Treat model choice as a sequential decision under uncertainty rather than a fixed pipeline: orchestrate several LLMs and spend more capable, more costly evaluation only where the decision is genuinely uncertain.

Stack

  • Multi-LLM orchestration
  • Bayesian decision theory
  • Python