AI Feature Attribution Stability Analyzer

Elara Intelligence

Hybrid · $160k – $200k / year · Full-time · Posted 4 months ago

This role is no longer accepting applications

Don't miss the next role like this

Create a free profile with your skills. We match you to new AI roles as they open, and companies hiring can contact you directly.

Create free profile, 2 minutes

Similar open roles you can apply to now:

See all open AI jobs

This role is no longer accepting applications

Don't miss the next role like this

Create a free profile with your skills. We match you to new AI roles as they open, and companies hiring can contact you directly.

Create free profile, 2 minutes

Similar open roles you can apply to now:

See all open AI jobs

About the role

A model explanation that changes significantly every time you ask it isn't trustworthy. SHAP values computed on the same instance twice, with only random seed variation, should be consistent. An attribution that says 'income was the top feature' one time and 'credit history was the top feature' another time for the same individual undermines confidence in the explanation and potentially violates regulatory requirements for consistent adverse action reasons. We build explainability tooling for 80 enterprise clients in financial services, healthcare, and hiring — domains where model explanations inform consequential decisions and face regulatory scrutiny. Feature attribution stability — measuring how consistent explanation outputs are across random seeds, minor perturbations, and repeated queries — is a critical quality dimension for explanation systems that most practitioners don't measure. You'll build the attribution stability analysis system: implementing stability measurement methodologies across SHAP, LIME, and gradient-based attribution methods, building the monitoring that tracks explanation consistency in production, and developing the reporting that helps clients understand which models and feature sets produce trustworthy explanations versus unreliable ones. The role is hybrid — 2 days per week in our NYC office.

Responsibilities

Design attribution stability measurement methodology across SHAP, LIME, and gradient-based methods
Build production monitoring tracking explanation consistency across repeated queries and minor input perturbations
Implement stability benchmarks comparing attribution methods on client model and data types
Create stability threshold recommendations tied to regulatory adverse action consistency requirements
Develop explanation quality reports surfacing instability patterns requiring model or feature attention
Research improved attribution methods with stronger theoretical stability guarantees

Requirements

6+ years in ML engineering with explainable AI, model interpretability, or fairness experience
Deep understanding of feature attribution methods and their theoretical properties
Strong Python and statistical analysis skills
Familiarity with regulatory explainability requirements in financial services or healthcare
Research track record in interpretable ML preferred