ML Infrastructure Engineer

TLDR:
We are looking for an ML Infrastructure Engineer to build the systems behind our LLM post‑training, RL, evaluation, inference, and agentic development workflows. You will work close to researchers, GPUs, training loops, data control systems, evals, inference stacks, and the infrastructure decisions that directly affect model learning and product quality.

White Circle is an AI Safety company building the safety, reliability, and optimization layer for AI systems. At the core of our platform are policies – simple natural‑language rules that define what an AI model should and shouldn’t do. We automatically test, enforce, and continuously improve these policies at scale.

• We’ve raised $11M from top funds, founders, and senior leaders at OpenAI, Anthropic, Hugging Face, Mistral, Deep Mind, Datadog, Sentry, and others
• We process over 100M+ API calls every month
• We fine‑tune and train our own LLMs so they run faster and cheaper than any open or proprietary model

We’re a small, highly focused team. If you want to work deeply on hard problems, see your work ship to production quickly, and influence how AI safety is actually built – you’re the one we need.

• Build robust, flexible, and scalable RL and post‑training pipelines, including smoke tuning runs for quality testing and approach ablations
• Design data control systems that govern what the model sees, when it sees it, and how training data flows through rollouts, replay, filtering, evaluation, and policy updates
• Tune training and inference end‑to‑end for high throughput across the systems that matter: networking, memory, compute scheduling, data loading, storage, checkpointing, and I/O
• Investigate how infrastructure choices affect learning dynamics, eval quality, model behavior, and training stability – staying close to the state of the art in LLMs, RL, and post‑training
• Build infrastructure for model iteration: experiment runs, artifacts, evals, dashboards, failure inspection, reproducibility, and cost visibility
• Work on inference infrastructure where it affects post‑training and evaluation loops
• Build and improve agentic development environments: coding‑agent harnesses, browser/tool integrations, terminal/runtime sandboxes, repo‑aware workflows, and multi‑agent orchestration
• Work closely with the team: plan future steps, discuss tradeoffs, share context early, and stay in touch while building

• Have designed, built, or maintained distributed RL/post‑training systems at scale and are fluent in their moving parts: rollouts, replay buffers, reward signals, data filtering, policy updates, evaluation loops, and failure analysis
• Are familiar with deep learning frameworks such as PyTorch or JAX
• Are proficient in Python, including concurrency, asynchronous programming, multiprocessing, and performance optimization
• Can debug distributed GPU workloads across CUDA runtime, container runtime, driver versions, NCCL or equivalent communication layers, networking, storage, scheduling, and checkpointing
• Have experience with profiling tools across the stack, for example py‑spy, PyTorch profiler, Nsight, perf, tracing, metrics, logs, or custom instrumentation
• Have experience with inference stacks such as vLLM, SGLang, TensorRT‑LLM, Dynamo, or custom serving infrastructure
• Can reason from system metrics back to model behavior: when latency, queueing, sampling, data order, rollout throughput, or infrastructure failures affect learning
• Have a strong ownership mindset: you can take an ambiguous infrastructure problem, make it concrete, ship a working system, and improve it from real feedback

• A public builder footprint: open‑source contributions to RL, distributed ML, LLM training, inference, eval, or agent infrastructure – repos, PRs, benchmarks, papers with code, technical posts – and a good technical X/Twitter presence with live building, debugging threads, and useful interaction with strong builders
• Experience in a high‑bar AI infra, research, or model environment such as xAI/Grok, Qwen, Byte Dance AI infra/research, Prime Intellect, or similar teams
• Custom training framework support or ownership: distributed training, fine‑tuning…