Machine Learning Engineer - High Profile Tech Hedge Fund

Machine Learning Engineer

Most AI infrastructure roles sit at one remove from the work that matters - maintaining platforms, managing queues, keeping the lights on. This is not that. We are a systematic trading firm where the research function runs on models, and where the gap between a capable system and a production-ready one is measured in outcomes. The engineer in this role sits at that gap.

We are not looking for someone to shepherd notebooks to deployment. We are looking for an engineer who understands both sides of the research-to-production boundary, who has opinions about inference architecture and the scars to back them up, and who wants to work somewhere the output of their work is immediately visible and unambiguously consequential.

What you will own

Research infrastructure - Build and maintain the systems that researchers depend on every day: training pipelines, experiment tracking, data tooling, and the evaluation frameworks that let quantitative teams iterate faster and with more confidence. Your instinct for what makes tooling good will shape how research gets done.

Inference and serving - Design and operate low-latency inference infrastructure for models that run in time-sensitive contexts. Understand the full stack — from model export and quantization to batching strategy and hardware utilization. Performance here is not a nice-to-have.

ML system design - Participate in architectural decisions about how models are trained, versioned, evaluated, and integrated into downstream systems. Work closely with researchers to translate experimental approaches into engineering requirements and then into working systems.

Technical environment

PyTorch / JAXCUDA / GPU optimization Distributed training Inference serving Python C++ / Rust Linux systems Large-scale data Model quantization Experiment infrastructure

The work day to day

• Design and ship systems that researchers trust with their most important work — models, experiments, data pipelines — and own them end to end
• Debug training instabilities, GPU memory bottlenecks, and latency anomalies with a level of rigor that treats correctness as non-negotiable
• Collaborate directly with quantitative researchers and ML practitioners; translate what they need into what can be built, and push back when the proposal won't hold
• Make real architectural decisions — training stack, serving strategy, evaluation design — and defend them with engineering reasoning, not consensus-seeking
• Write code that is maintained by people with very high standards and that will be depended upon in production

Who you are

• Deep experience with ML systems in production - not just training pipelines you inherited, but systems you designed, broke, and improved
• Fluency at the boundary of ML and systems engineering: you understand how GPU memory hierarchies affect training throughput, how batching strategies affect tail latency, and why these things matter
• A background at a serious organization - a frontier AI lab, a quantitative firm, a top-tier technology company - where engineering standards were genuinely high and you were shaped by them
• Evidence of having built something hard: not contributed to, not maintained, but driven through from prototype to production with your fingerprints on the design decisions
• Intellectual curiosity about the domain this infrastructure supports — you don't need a finance background, but you should find it interesting that your work connects directly to measurable outcomes
• A preference for small, senior teams over large organizations - you want to know the people you work with and be accountable to them
• BS / MS / PhD in Computer Science, Electrical Engineering, or a related field - or equivalent demonstrated by the work

Why this role

AI infrastructure roles at trading firms rarely appear publicly, and when they do, the description usually undersells the problem difficulty. The models here are not demos. The infrastructure here is not a portfolio project. The feedback loop between a system performing well and an outcome that is measurable is short. If you have spent your career building ML systems that demanded your best, and you want to bring that somewhere where the stakes are real, this is worth a conversation.