AI LLM Output Guardrail Framework Developer

Yovex Intelligence

Remote · $130k – $160k / year · Full-time · Posted 5 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

Prompt injection is the input side of LLM security. Output guardrails are the other half — ensuring that model responses don't leak private training data, generate harmful content, produce hallucinated citations, or violate enterprise content policies. Our guardrail framework evaluates every LLM response in real time and blocks or modifies non-compliant outputs before they reach the user. The framework runs a battery of classifiers on each response: PII detection, hallucination scoring, toxicity filtering, brand safety assessment, and custom policy checks defined by each enterprise client. The latency budget is 100ms total — all checks must complete before the response streams to the user. You'll build the classifier ensemble, the policy engine that orchestrates them, and the response modification pipeline that fixes minor violations without blocking the entire response. The LLM security team is five engineers. You'll be the senior engineer who designs the guardrail architecture, balancing safety thoroughness against the latency constraints of real-time response streaming.

Fully remote.

Responsibilities

Design the output guardrail classifier ensemble for real-time evaluation
Build PII detection, hallucination scoring, and toxicity filtering classifiers
Implement custom policy engines for enterprise-specific content rules
Create response modification pipeline fixing minor violations inline
Optimize ensemble latency to stay within 100ms total budget
Evaluate guardrail effectiveness against adversarial test suites

Requirements

5+ years in NLP or ML security with production classification experience
Experience building real-time classifier pipelines with strict latency constraints
Strong Python and PyTorch skills for model development
Understanding of LLM security threats: data leakage, hallucination, toxicity
Track record of shipping content safety features