Principal Research Scientist, Synthetic Data Generation

Details

NVIDIA is seeking a Principal Research Scientist to set the technical direction for synthetic data generation across its frontier model efforts. The role involves defining and building open-source libraries within the NVIDIA NeMo ecosystem to generate synthetic datasets across text, code, structured, and multimodal data for pre-training and post-training large language models such as Nemotron. This position combines hands-on software engineering with applied research in generative methods and involves collaboration with research, engineering, product, model teams, and external labs.

Responsibilities

• Build and scale data-generation pipelines using LLM-based methods and automated quality evaluation for datasets supporting the initial training and fine-tuning of LLMs such as Nemotron.
• Develop pipelines covering reasoning, coding, structured output, and multimodal understanding.
• Pioneer data generation for agentic and tool-use training, including synthetic trajectories, multi-turn interactions, function calling, executable reinforcement-learning environments, reward modeling, and verifiable-reward data.
• Advance multimodal synthetic data generation for images, documents, video, and audio in partnership with NVIDIA model teams.
• Advance privacy-preserving and safe synthesis using differential privacy, anonymization, and de-identification for model training on sensitive data in regulated domains.
• Develop and maintain open-source libraries and SDKs with clean APIs and strong documentation.
• Drive software excellence through modern tooling, configuration-based architecture, and professional Git and CI/CD practices.
• Publish original research at leading machine learning and AI conferences.
• Mentor scientists and engineers across the team.

Requirements

• PhD in Computer Science, Machine Learning, Statistics, or a related field, or equivalent experience.
• 15+ years of engineering and research experience in synthetic data generation, generative modeling, multimodal machine learning, or related areas.
• Deep technical understanding of LLMs and how data influences pre-training, post-training, and reinforcement-learning stages.
• Experience with inference frameworks such as vLLM or TGI.
• Proven track record of developing or maintaining software libraries used by a broad developer community.
• Experience building and optimizing scalable data pipelines for large-scale model training, including throughput, distributed inference, and cluster-scale cost optimization.
• Strong publication record at premier venues such as NeurIPS, ICML, ICLR, ACL, or similar.

Preferred Qualifications

• Significant open-source contributions in machine learning or data tooling with community adoption.
• Experience with multimodal generation or understanding, including vision-language, document AI, video, or audio.
• Experience generating data for agentic, tool-use, or reinforcement-learning post-training, including reinforcement-learning environment design.
• Background in differential privacy, de-identification, or synthetic data for regulated industries such as healthcare, finance, or government.
• Experience influencing model-training decisions at frontier scale or partnering directly with pre-training and post-training teams.

Benefits

• Equity and NVIDIA benefits.
• NVIDIA is committed to fostering an inclusive work environment and is an equal opportunity employer.