AI/Synthetic Engineer

SHRM•Alexandria, VA

13d•Hybrid

About The Position

Build the organization’s synthetic-audience capability: the data-grounding layer that connects first-party audience data to large language models, the synthetic-audience models themselves, and the engineering pipelines and evaluation harness around them. Direct the synthetic-capability partner and the foundation-model infrastructure beneath it and operate the validation harness that the team will use to judge accuracy. This is an engineering role at heart but applied entirely to research questions.

Requirements

Bachelor’s degree in computer science or related field or relevant equivalent experience in lieu of degree.
Seven (7) or more years in ML/LLM engineering or applied data science, including production systems.
Hands-on experience with LLM application development: grounding/RAG, prompt engineering, and evaluation frameworks.
Strong data engineering: pipelines, embeddings, working with first-party datasets.
Understanding of model evaluation and the failure modes of generative systems.
Able to work to research-defined validation standards rather than ship unchecked.
Ability to effectively leverage artificial intelligence (AI) tools and technologies to streamline workflows, enhance productivity, and improve overall work quality.

Nice To Haves

Master’s degree preferred.
Exposure to synthetic data, agent-based simulation, or survey/behavioral data preferred.
Experience directing an AI/ML vendor or platform partner preferred.

Responsibilities

Design and build the first-party data grounding layer (embeddings / retrieval) that anchors synthetic audiences in real segment data.
Build, tune, and maintain synthetic-audience models and the prompt/evaluation pipelines around them.
Operate the validation harness: run synthetic output against real holdout data and surface accuracy bounds for the Verification & Validation Data Scientist to interpret.
Direct the synthetic-capability partner; manage foundation-model infrastructure and keep the architecture portable across providers.
Build guardrails that prevent confident-but-wrong output from reaching stakeholders unflagged.
Protect data ownership and ensure member data never trains external shared models.