About this role
By joining Sedgwick, you'll be part of something truly meaningful. It’s what our 33,000 colleagues do every day for people around the world who are facing the unexpected. We invite you to grow your career with us, experience our caring culture, and enjoy work-life balance. Here, there’s no limit to what you can achieve.
Newsweek Recognizes Sedgwick as America’s Greatest Workplaces National Top Companies
Certified as a Great Place to Work®
Fortune Best Workplaces in Financial Services & Insurance
AI Evaluation Engineer
PRIMARY PURPOSE: The AI Evaluation Engineer designs, builds, and operates the evaluation systems that determine whether our GenAI solutions are accurate, safe, reliable, and ready for production. You will serve multiple product pods, establishing consistent evaluation standards, automated pipelines, and quality gates, and you will help teams measure and improve LLM, RAG, and agentic applications throughout their lifecycle.
Fully remote role, open to U.S. based candidates.
ESSENTIAL FUNCTIONS AND RESPONSIBILITIES
- Establishes standardized, automated evaluation frameworks for LLMs and traditional ML models, including LLM as a judge patterns, ROUGE, METEOR, and statistical validation to ensure performance, reliability, and readiness prior to production release.
- Deploys comprehensive observability solutions to monitor model drift, hallucination risk, latency, throughput, and token consumption, balancing model quality with infrastructure cost and operational efficiency.
- Define evaluation methodologies and metrics for LLM, RAG, and agent-based applications (e.g., faithfulness, groundedness, relevance, task completion, tool-use correctness, safety).
- Build reusable evaluation frameworks, templates, and libraries that pods can adopt with minimal effort.
- Partner with product owners and SMEs to translate business requirements into measurable quality criteria and acceptance thresholds.
- Develop automated evaluation pipelines integrated into CI/CD, with regression testing and quality gates for prompt, model, and retrieval changes.
- Implement offline evaluation (benchmark and golden datasets) and online evaluation (production monitoring, sampling, drift and quality alerts).
- Support A/B and shadow testing for model, prompt, and architecture changes.
- Design, validate, and maintain LLM-as-a-judge evaluators, including rubric design, bias mitigation (position, verbosity, self-preference), and calibration against human labels.
- Establish human annotation workflows, labeling guidelines, and inter-annotator agreement practices with SMEs.
- Track judge reliability over time and re-calibrate as models and use cases change.
- Build and curate golden datasets, adversarial and edge-case suites, and synthetic test data generation approaches.
- Apply approved data-handling practices (de-identification, synthetic data, access controls) for any dataset derived from sensitive or regulated data.
- Implement red-teaming, hallucination, toxicity, PII leakage, prompt-injection, and jailbreak testing.
- Produce evaluation evidence and reporting that supports security, privacy, compliance, and model risk reviews.
- Contribute to responsible AI standards and release-readiness criteria.
- Embed with pods to onboard them to evaluation practices, then coach teams toward ownership of their own eval suites.
- Build dashboards and reporting that give engineering and business leaders transparent quality and cost/latency trends.
- Evaluate and recommend eval tooling and platforms; document standards and best practices.
ADDITIONAL FUNCTIONS and RESPONSIBILITIES
- Performs other duties as assigned.
- Travel as required.
QUALIFICATIONS
Education & Licensing
Master’s degree in a computer science or a quantitative discipline from an accredited college or university preferred.
Experience
4+ years in software, ML, or data engineering, with at least 1-2 years working hands-on with LLM-based applications (adjust to your market).
Skills & Knowledge
- Professional experience with evaluation and observability frameworks like LangSmith, Arize Phoenix, or WhyLabs
- A strong understanding of statistical validation, model evaluation metrics, and the ability to partner with Data Scientists to optimize model performance
- Strong Python skills and software engineering fundamentals (testing, version control, CI/CD, code review).
- Practical experience evaluating LLM/RAG systems, including retrieval metrics (recall, precision, MRR/nDCG) and generation metrics.
- Hands-on experience with LLM-as-a-judge design, prompt engineering, and judge validation.
- Working knowledge of statistics and experimental design: sampling, confidence intervals, significance testing, inter-rater agreement.
- Familiarity with at least one eval or observability tool (e.g., Ragas, DeepEval, promptfoo, LangSmith, Langfuse, MLflow, or equivalent).
- Experience with cloud platforms (Azure, AWS, or GCP) and their managed LLM services.
- Clear communication skills and the ability to work with engineers, product owners, SMEs, and risk stakeholders.
- Excellent oral and written communication skills, including presentation skills
- Ability to work in a team environment
- Ability to meet or exceed Performance Competencies
Preferred
- Experience evaluating agentic systems (multi-step reasoning, tool calling, trajectory evaluation).
- Background in ML evaluation, search/relevance, NLP, or QA automation.
- Experience with LLMOps, monitoring and observability, and production feedback loops.
- Familiarity with AI risk and governance frameworks (e.g., NIST AI RMF, ISO/IEC 42001) and secure development practices in regulated environments.
- Experience with data annotation workflows and working with SMEs.
- Knowledge of enterprise controls: SSO, RBAC, private networking, logging, and encryption.
SUCCESS MEASURES (first 12 months)
- Standard evaluation framework adopted by pods.
- Automated quality gates in CI/CD for production-bound solutions.
- Calibrated LLM judges with documented agreement against human labels.
- Reduction in regressions reaching production and in time to validate model or prompt changes.
WORK ENVIRONMENT
When applicable and appropriate, consideration will be given to reasonable accommodations.
Mental: Clear and conceptual thinking ability; excellent judgment, troubleshooting, problem solving, analysis, and discretion; ability to handle work-related stress; ability to handle multiple priorities simultaneously; and ability to meet deadlines
Physical: Computer keyboarding, travel as required
Auditory/Visual: Hearing, vision and talking
The statements contained in this document are intended to describe the general nature and level of work being performed by a colleague assigned to this description. They are not intended to constitute a comprehensive list of functions, duties, or local variances. Management retains the discretion to add or to change the duties of the position at any time.
Sedgwick is an Equal Opportunity Employer and a Drug-Free Workplace.