Senior Research Engineer/Scientist

ServiceNow•Montreal, QC

1d•Hybrid

About The Position

The Core AI Model Training team leads research and engineering across the full lifecycle of large language model development for enterprise use cases: data curation, pretraining, fine-tuning, and evaluation. Our goal is to continuously improve our custom enterprise models by incorporating novel techniques and optimizing training and inference efficiency, establishing a differentiated advantage for AI applications across the platform. These models are consumed across multiple business units to power a wide range of use cases, enabling teams to solve complex problems and deliver tailored solutions. You'll help build the next generation of enterprise language models that bring AI experiences into our customers' day-to-day work, serving 9K+ enterprise customers worldwide. We're just getting started with our early-adopter customers, and we need your help building an amazing range of solutions.

Requirements

Expertise in LLM post-training, including supervised fine-tuning and distillation.
Expertise in reinforcement learning for LLMs, including PPO, GRPO, DPO, and reward modeling.
Hands-on experience with training frameworks and distributed/large-scale training, including FSDP, DeepSpeed, and Megatron, as well as parallelism strategies such as tensor, pipeline, expert, and data parallelism.
Experience with synthetic data generation for training and evaluation.
Experience with data curation at scale, including deduplication, filtering, and data-mixture design.
Familiarity with a range of transformer architectures, including decoder-only/autoregressive, encoder-decoder, and mixture-of-experts.
Expert-level Python with strong OOP and design-pattern fundamentals.
Ability to read current research and rapidly prototype and experiment with new ideas.
6+ years of relevant experience with a Bachelor's degree, 4+ years with a Master's degree, a PhD, or equivalent experience.

Nice To Haves

Grounding in evaluation and benchmarking, including building evaluation harnesses, contamination checks, and custom enterprise benchmarks.
Pretraining experience or an end-to-end perspective across pretraining and post-training, with an understanding of how choices in one stage propagate to the other.
Familiarity with inference and serving technologies such as vLLM or SGLang, quantization, and KV-cache tradeoffs.
Publications at top-tier venues such as ICLR, NeurIPS, ICML, ACL, EMNLP, or AAAI.
Fluency with AI productivity tools such as Claude Code and Codex, with a track record of thoughtfully integrating AI into engineering and research workflows

Responsibilities

Confronted with real-world challenges and datasets, you will use your AI/ML expertise and creativity to apply existing methods and develop new ones to solve problems in a practical and scalable way.
Research, propose, implement, train, and evaluate models and techniques end to end.
Build and maintain training pipelines and evaluation harnesses that make results reproducible and progress measurable.
Collaborate daily with research scientists, engineers, and product managers to ship high-quality, high-impact work.
Own your work from design through implementation, testing, and delivery, partnering with product owners to translate requirements into results