Staff ML Systems Engineer – Explainability

Elloe AI; Austin (TX); Hybrid; USD 180,000 - 260,000; Full time

Benefits offered by this job

Real equity

Remote-first

Job summary

Elloe AI is the trust layer for AI, and this Staff ML Systems Engineer role owns the explainability stack used to inform deployment decisions. You’ll build SHAP overlays that run in production with sub-100ms latency and real trace logs, tied to actual decisions.

We expect you to ship ML infrastructure at real-world scale, reason about SHAP limits, and collaborate across policy, risk, and product to ensure compliance and auditable outputs.

Qualifications

• Shipped ML infra at scale with real latency targets.
• Knowledge of SHAP and limitations, with practical workarounds.
• Ability to derive production-ready explainability from first principles.
• Experience explaining models to non-technical stakeholders.

Responsibilities

• Own the explainability stack used for production deployment decisions.
• Build SHAP overlays that run inline with model responses and are auditable.
• Integrate explainers across Claude, GPT-4 and Mistral with version control.
• Create dashboards, audit logs, and human-readable traces for compliance.

Skills

SHAP explainability

ML infra

Production latency

Retrieval explainability

Vector search

Tools

Claude

GPT-4

Mistral

Job description

Staff ML Systems Engineer - Explainability at Elloe AI

Full-time | Remote | ML Infrastructure | Reports to CTO

About Elloe

Elloe is the trust layer for AI.

We sit between the world’s most powerful language models and the institutions that can't afford to get it wrong - hospitals, banks, regulators. We trace and block failures in real time. That's not marketing - we're deployed at the European Commission, with NIH clinical trials, and inside a Top-5 EU bank catching GDPR violations live.

This is the enforcement layer GenAI has been missing. We're not visualizing problems - we're fixing them.

About the Role

This role owns the explainability stack that institutions will base deployment decisions on. You'll build SHAP overlays that don't just run in notebooks - they run in production, under 100ms, with real trace logs, tied to real decisions.

It's fast, it's visible, and it has to hold up in court.

What You'll Build

1. SHAP + Retrieval Explainability

• Build attribution for both retrieval and generation paths
• Design SHAP-based systems that can run inline with model responses
• Handle vector search + hallucination tracing, not just token salience

2. Infrastructure for Real Users

• Integrate explainers across Claude, GPT-4, Mistral - with version control and adapter logic
• Build APIs that feed into dashboards, audit logs, and human-readable traces
• No toy demos. Every output has to stand up to legal review

3. Enforcement Hooks

• Align your outputs with EU AI Act Article 10 / 14 requirements
• Feed explainability directly into policy engines, risk graphs, and compliance frontends
• Help product and sales use your stack as a reason to buy, not a nice-to-have

Who You Are

• You've shipped ML infra - real-world scale, real latency targets
• You know the limits of SHAP and how to work around them
• You don't wait for product specs - you work from first principles
• Bonus: You've had to explain a model to someone who doesn't trust it (a lawyer, a doctor, an auditor)

Why It Matters

Every buyer we talk to - from governments to hospitals - wants explainability. But most of what's out there doesn't scale or doesn't mean anything to the people making the actual decisions.

This role builds the version that matters. You won't just make models legible - you'll help make them safe to use.

Logistics & Application

• Start Date: Flexible (Q3 ideal)
• Location: Remote-first, timezone overlap with NY or EU preferred
• Comp: Top of market salary + real equity