As a Research Engineer on the Frontier Foundations team, you will partner with Research Scientists to build the systems that run large-scale LLM training, reinforcement learning, and inference at Datadog. You will build and operate an in-house distributed runtime that schedules work across large GPU clusters, and you will own how quickly researchers can iterate and how reliably long training runs complete.
Building on our track record of AI-powered solutions (e.g., Bits AI, Bits Evolve, and our time series foundation model Toto), Datadog AI Research (DAIR) tackles high-risk, high-reward problems grounded in real-world challenges in cloud observability and security.
The Frontier Foundations team builds and maintains the high-performance infrastructure that lets us train and serve large language models at scale. It supports both of our research areas:
Trained Agents for Observability -- Post-training models to operate autonomously across Datadog's domain. SRE incident response is our first target, with a clear path to code repair, security response, and infrastructure optimization. We build the simulation environments, RL training loops, and evaluation infrastructure needed to train agents that match or surpass frontier models at a fraction of the cost.
What You'll Do:
Design, build, and operate distributed training and inference infrastructure for large language models, including Mixture-of-Experts models
Scale training with parallelism strategies (data, tensor, pipeline, and expert), memory optimization, checkpointing, and failure recovery
Build inference and rollout systems that support evaluation, experimentation, and RL without slowing down training
Build execution environments and sandboxes for online RL and agent post-training
Own continuous performance validation: regular benchmarks of rollout speed, training throughput, and learning behavior, with dashboards that make regressions visible
Improve the researcher experience: job submission, scheduling, quotas, GPU utilization, and observability
Work with platform teams on cluster capacity, networking, and scheduling across GPU providers
Investigate performance and reliability problems across hardware, software, and distributed systems
Who You Are:
You have depth in distributed systems and ML infrastructure for training and inference at scale; experience with Ray, Kubernetes, Slurm, or similar frameworks is a plus
You are proficient in Python and in at least one systems language (e.g., Rust, C++, or Go) ● You have experience operating large-scale training or serving infrastructure, or schedulers and orchestration systems for GPU or other high-performance workloads
You understand what limits throughput/latency in distributed inference workloads: communication, memory, scheduling, and stragglers
You have strong debugging, profiling, and incident-diagnosis skills, ideally including on-call experience
You can explain design and performance trade-offs clearly to both technical and non-technical audiences
Prior experience training machine learning models is valuable but not essential for engineers with strong distributed systems or HPC backgrounds
Bonus Points (any of the following):
You are familiar with distributed training frameworks such as Megatron-LM, TorchTitan, or DeepSpeed
You have experience with LLM inference engines such as vLLM or SGLang, including batching and KV-cache management
You have experience with RL training systems and rollout infrastructure
You have experience with collective communication (NCCL), RDMA, or low-latency networking
You have experience with GPU benchmarking, profiling, and cluster performance monitoring
You have experience building sandboxed code execution or large-scale evaluation infrastructure
You have contributed to open-source ML or distributed systems projects
Datadog values people from all walks of life. We understand not everyone will meet all the above qualifications on day one. That's okay. If you're passionate about technology and…