Software Engineer - Machine Learning Technology - Marquee Hedge Fund

The problemMost ML infrastructure roles exist inside large organizations where the feedback loop between your work and something real is long, indirect, and easily rationalized away. You build a platform, someone uses it, a metric moves somewhere, a VP is pleased. The connection between what you built and what it produced is theoretical at best.This role however is with a systematic trading firm where models are the business, where inference latency is measured in consequences, and where the distance between a system performing well and an outcome that registers is as short as it gets in this industry. If you have spent your career at a place where engineering standards were genuinely uncompromising - where code review meant something, where performance regressions were treated as incidents, where "good enough" was not an accepted answer - this is written for you.What you will ownResearch infrastructure The systems that let quantitative researchers move fast without breaking things: training pipelines, experiment tracking, evaluation frameworks, and data tooling built to the standard of people who will immediately notice when it isn't. Your judgment about what good tooling looks like will shape how research gets done here - not in a committee, not through a roadmap process, but directly.Inference and serving Low-latency inference infrastructure for models that operate in time-sensitive contexts. You understand the full stack - model export, quantization, batching strategy, hardware utilization - and you have strong opinions about the tradeoffs because you have made them before under real constraints. Performance here is load-bearing.ML system design Architectural decisions about how models are trained, versioned, evaluated, and integrated into production. You will work directly with quantitative researchers, translate experimental requirements into engineering designs, push back when a proposal won't hold, and own the outcome when it ships.The technical environmentPyTorch / JAX — CUDA / GPU optimization — Distributed training — Inference serving — Python — C++ / Rust — Linux systems — Large-scale data pipelines — Model quantization — Experiment infrastructureDay to dayDesign and ship systems that the research team trusts with their most important work, and own them fully - including when they breakDebug training instabilities, GPU memory bottlenecks, and latency anomalies with the rigor of someone who treats correctness as a hard constraint, not a targetMake real architectural decisions — training stack, serving strategy, evaluation design — and defend them with engineering reasoningWrite code that will be read by engineers with very high standards and depended upon in production, indefinitelyWork directly with quantitative researchers and ML practitioners as a peer, not a service providerWho you areYou have built ML systems in production at a place where the bar was genuinely high - a frontier AI lab, a top-tier technology company, a well-capitalized startup where the engineering culture was set by people who cared deeply about craft. You were not just present for the work. Your fingerprints are on the design decisions.You operate fluently at the intersection of ML and systems engineering. You understand why GPU memory hierarchies affect training throughput, how batching strategy affects tail latency under load, and what it costs to get these things wrong. These are not things you have read about.You have driven something hard from prototype to production - not contributed to it, not maintained it after someone else built it, but conceived of it, argued for the design, and shipped it into an environment where it was depended upon.You prefer small, senior teams to large organizations. You want to know the people you work with. You want to be accountable to them and have them be accountable to you. You find the idea of your infrastructure connecting directly to measurable outcomes more motivating than the alternative.A BS, MS, or PhD in Computer Science, Electrical Engineering, or a related field — or a body of work that makes the credential irrelevant.Why this is worth your attentionRoles like this do not appear often. The serious ML infrastructure problems in systematic trading are kept close, the teams are small, and the openings are rare. What is on offer here is not a compensation package or a brand name - it is a set of problems that will demand your best, a team of people who will meet that standard alongside you, and a feedback loop between your work and something real that is shorter than almost anywhere else you could go.If your current role has started to feel too large, too slow, or too disconnected from outcomes that matter - this is worth a conversation.