OBSIDIAN LENS / AI BIAS DETECTION PLATFORM

Is your AI treating everyone fairly?

ETHYX AI helps teams inspect datasets and automated decisions for hidden unfairness before they affect real people.

FAIRNESS RISK SCORE

67

MEDIUM RISK
Disparate Impact
0.74
FAIL
Equal Opportunity
-0.18
FAIL
Calibration
0.92
PASS

500

Rows Analyzed

4

Fairness Checks

3

Risks Detected

1

Mitigation Plan

THE PROBLEM

AI bias is invisible — until it harms someone

ETHYX flags hidden bias before your model ever reaches production.

Historical Discrimination

AI models trained on biased data repeat and amplify past inequalities at scale.

High-Stakes Decisions

Hiring, lending, and insurance algorithms need fairness checks before deployment.

Healthcare & Safety

Biased diagnostic models can deny care to entire populations without anyone noticing.

CORE CAPABILITIES

Everything you need to audit AI fairness

DATASET BIAS SCAN

Upload any tabular dataset and detect statistical imbalances across protected attributes like gender, race, and age.

FAIRNESS METRICS

Compute Disparate Impact, Statistical Parity, Equal Opportunity, and Calibration with one click.

GROUP IMPACT COMPARISON

Visualize how model outcomes differ across demographic subgroups with interactive charts.

EXPLAINABLE DECISIONS

AI-powered plain-English explanations of why each metric passed or failed and what it means.

MITIGATION STRATEGY

Receive actionable recommendations to rebalance training data or adjust model thresholds.

AUDIT REPORT EXPORT

Generate a comprehensive PDF fairness report ready for compliance review and stakeholder sharing.

WORKFLOW PIPELINE

Four Steps to Fair AI

01

Upload Dataset

Upload your CSV or connect to your data warehouse. ETHYX supports datasets up to 500k rows.

02

Map Attributes

Select your target variable and protected attributes like gender, race, or age group.

03

Run Fairness Audit

ETHYX computes five key fairness metrics and generates AI-powered explanations in seconds.

04

Review & Mitigate

Explore results, apply mitigation strategies, and export a compliance-ready audit report.

FAIRNESS METRICS

Five Pillars of Fair AI

ETHYX computes these industry-standard metrics on every audit to give you a complete picture of model fairness.

Disparate Impact

P(Y=1|D=0) / P(Y=1|D=1)

Threshold: ≥ 0.80

Ratio of positive outcome rates between unprivileged and privileged groups. Below 0.80 signals adverse impact.

Statistical Parity

P(Ŷ=1|D=0) - P(Ŷ=1|D=1)

Threshold: ± 0.05

Difference in selection rates. Measures whether the model selects at equal rates regardless of group membership.

Equal Opportunity

TPR(D=0) - TPR(D=1)

Threshold: ± 0.05

Difference in true positive rates. Ensures qualified individuals have equal chance of positive prediction.

Average Odds

½(ΔFPR + ΔTPR)

Threshold: ± 0.05

Average of FPR and TPR differences. A balanced measure of equalized odds across groups.

Calibration

E[Y|Ŷ=p, D=d] ≈ p

Threshold: > 0.90

Whether predicted probabilities match observed outcomes equally for each group. Key for risk scoring.

DEMO REPORT

See a Real Audit Report

MEDIUM FAIRNESS RISK
Score: 67
MetricValueThresholdResult
Disparate Impact0.740.80
FAIL
Statistical Parity-0.24±0.05
FAIL
Equal Opportunity-0.18±0.05
FAIL
Calibration0.92>0.90
PASS

Build fairer AI before launch.

Join the teams using ETHYX AI to ship responsible machine learning models with confidence.