ML Engineer - LLM & Agentic Evaluations at Turing
posted 3 hours agoML Engineer – LLM & Agentic Evaluations | Remote (LATAM) | Contract, 3 months
About Turing
Turing is a San Francisco-based research accelerator for frontier AI labs and a trusted partner for global enterprises deploying advanced AI systems.
Role Overview
Join Turing's AI Evaluation team to design and execute evaluations for Large Language Models (LLMs) and agentic AI workflows. This is a hands-on, research-flavored role: you will work alongside ML engineers, researchers, and domain experts on real evaluation problems that directly influence how leading AI models are trained, refined, and shipped.
What You'll Do
- Design, build, and run evaluation suites for LLMs across reasoning, coding, math, tool use, and long-context tasks.
- Develop benchmarks and rubrics for agentic workflows, including multi-step planning, tool/function calling, code execution, and browser/computer-use agents.
- Curate, clean, and annotate evaluation datasets; write clear task specifications and grading criteria.
- Run model comparisons (open-source and frontier models), analyze results, and produce structured reports with actionable insights.
- Implement automated graders (programmatic checks, LLM-as-a-judge) and validate them against human reference judgments.
- Investigate failure cases, characterize error patterns, and write reproducible findings for the research team.
- Collaborate with engineers to integrate evaluations into internal harnesses, CI pipelines, and dashboards; contribute to documentation, best practices, and tooling.
Requirements
- Currently pursuing or recently completed a Bachelor's, Master's, or PhD in Computer Science, Machine Learning, Data Science, Mathematics, or a related quantitative field.
- Strong programming skills in Python, including the standard data and ML stack (NumPy, Pandas, Jupyter).
- Working understanding of how modern LLMs are trained, prompted, and evaluated.
- Experience interacting with LLM APIs (OpenAI, Anthropic, Google, open-source via Hugging Face / vLLM).
- Strong analytical and writing skills — able to clearly document tasks, results, and recommendations.
- Comfortable working independently in a fully remote environment with overlapping collaboration hours.
Nice to Have
- Experience designing or running LLM evaluations or benchmarks (e.g., MMLU, GPQA, SWE-bench, GAIA, AgentBench, BIG-bench, HumanEval, TerminalBench, WebArena).
- Hands-on experience with agent frameworks (LangGraph, AutoGen, CrewAI, OpenAI Agents SDK) or building tool-using agents from scratch.
- Familiarity with LLM-as-a-judge methodologies, inter-annotator agreement, and statistical analysis of model outputs.
- Experience with prompt engineering, few-shot design, and structured output formats (JSON Schema, function calling).
- Exposure to RLHF/RLAIF, fine-tuning, or post-training data pipelines.
- Background in software engineering, data annotation tooling, or NLP/ML research publications.
Offer Details
- Engagement type: Contractor assignment (no medical/paid leave).
- Duration: 3 months, adjustable based on engagement.
- Commitment: Minimum 20 hours per week, with at least 5 hours/day of overlap with PST.
- Location: Remote (LATAM).
- Hiring process: Technical interview.