J.U.P.I.T.E.R.
An AI auditing system for HR bias detection in federal contracting pipelines — built to give compliance officers real-time visibility into algorithmic hiring decisions.
The Problem
Federal contractors subject to OFCCP oversight spend hundreds of hours per quarter documenting that their ATS-driven hiring decisions don't produce disparate impact for protected classes. The problem: the audit tooling is 15 years old, the reporting formats don't match what regulators actually request, and most compliance officers have no visibility into the algorithmic layers making candidate ranking decisions.
J.U.P.I.T.E.R. (Justice · Unbiased · Processing · Intelligence · Technology · Ethics · Regulation) was designed to make the black box transparent — and auditable.
System Architecture
- Ingestion Layer: ATS data connectors (Workday, Greenhouse, Lever) pipe application events into a standardized audit schema.
- Bias Detection Engine: Statistical models flag adverse impact across 8 demographic dimensions; results surfaced in real-time as candidate pools are filtered.
- Explainability Layer: Each algorithmic decision is logged with a feature-importance breakdown — what signals drove the ranking, with what weight.
- Audit Export: One-click OFCCP-formatted reports generated from the audit log — no manual reformatting required.
Audit Dashboard UI
The compliance officer-facing dashboard was designed around one principle: a non-technical user should be able to understand what the algorithm did and why within 90 seconds of opening any report.
The interface uses a traffic-light risk scoring system (Green / Amber / Red) at the role level, drills into demographic breakdown charts for flagged roles, and surfaces the top 3 algorithmic factors driving any bias signal — each with a plain-language explanation generated by GPT-4.
Ethics Framework
Beyond the technical system, J.U.P.I.T.E.R. ships with a configurable ethics policy layer — organizations set their own thresholds for what constitutes a reportable adverse impact signal, with an audit trail of every threshold change (who changed it, when, and why).
This was a deliberate design choice: the tool doesn't impose a single definition of fairness — it makes the organization's chosen definition explicit, auditable, and defensible under regulatory scrutiny.
Government Use Case
- Piloted with 3 federal contractors subject to OFCCP oversight — $50M+ in combined annual contract value.
- Compliance review time reduced by 70% — from 12 days per quarterly audit to 3.5 days.
- Zero OFCCP findings against pilot organizations in the post-implementation audit cycle.
- Explainability reports accepted directly by regional OFCCP offices as primary audit documentation — first tool to achieve this in the pilot program.