Senior AI Research Scientist for Enterprise Agents

Servicenow; Santa Clara (CA); On-site; USD 213,000 - 354,000; Full time

Benefits offered by this job

Health plans

Flexible spending accounts

401(k) Plan with company match

ESPP

Matching donations

Flexible time away plan

Family leave programs

Job summary

ServiceNow seeks a Staff Research Scientist to lead a major workstream in agent learning and recursive self-improvement. You will convert failures and successful trajectories into hypotheses, experiments, training signals, and deployable model improvements, across reasoning, memory, tool use, and multi-agent coordination.

This research role blends scientific reasoning, training code, agent systems, and production constraints, with collaboration across product, engineering, infrastructure,

Qualifications

• 7+ years of ML/AI research or engineering experience; a PhD or advanced degree is required.
• Strong foundations in ML/DL, RL, and probabilistic experimentation with hands-on LM or multimodal model work.
• Ability to take ambiguous questions from hypothesis through implementation, experiment, analysis, and measurable outcomes.
• Experience with distributed training, rollout, and inference using PyTorch distributed/FSDP, DeepSpeed or Megatron; and post-training systems.

Responsibilities

• Design and execute end-to-end research projects to improve enterprise agents in planning, memory, tool use, and verification.
• Research model post-training methods: continued pretraining, SFT, RL, reward modeling, and distillation.
• Build reproducible training environments and pipelines for training, evaluation, and deployment; collaborate with cross-disciplinary teams.

Skills

Python

PyTorch

Machine learning

Reinforcement learning

Multimodal models

Large language models

Distributed training

Research experience

Education

PhD or advanced degree

Tools

DeepSpeed

Megatron

Ray

Kubernetes

vLLM

SGLang

Job description

ServiceNow seeks a Staff Research Scientist to lead a major workstream in agent learning and recursive self-improvement. You will convert failures and successful trajectories into hypotheses, experiments, training signals, and deployable model improvements, across reasoning, memory, tool use, and multi-agent coordination.

This research role blends scientific reasoning, training code, agent systems, and production constraints, with collaboration across product, engineering, infrastructure,