Deep Learning Technical Staff

Citadel Securities is one of the largest market makers in the world, providing liquidity across nearly every major asset class and market globally, in real time. Machine learning is fundamental to how we do it. We are the firm’s centralized applied-AI team, working with strategy teams across the firm. We conduct the frontier model research that benefits all of them in common — the deep-learning models that predict where prices are going — and we tailor that research to the highest-impact strategies. We build the systems that turn those predictions into trading decisions, as well as the tooling that enables researchers across the firm to apply machine learning to their own problems. As the centralized team, we take on the hardest applied-AI problems first and set the standard the wider organization follows. Our research is applied: it ships to production and is measured by its impact on the business. Every member of the team is a member of technical staff — a strong generalist with depth in at least one specialty, whether deep learning, large-scale ML systems, applied LLMs and agents, or the mechanics of markets themselves. We operate in small teams with end-to-end ownership and minimal bureaucracy. The role You build ML applications and systems that move real money, and you want to understand why they work — not simply ship them. You write code once you understand the research context and why it matters, and you are equally comfortable reading a research paper and profiling a training run. As Quantitative AI Technical Staff, you will have substantial independence to pursue the research directions you believe are most impactful, supported by state-of-the-art clusters with a very high compute-to-researcher ratio and by dedicated engineering, hardware, and systems teams that help bring your ideas into production. You will be responsible for improving every part of our models — from how we featurize raw market data, to architecture design, to training dynamics, to how a model’s predictions become trading decisions. The role is approximately half research and half engineering. Your work will be directly and measurably impactful on the business, and it will be challenging: this is a field with no easy or obvious solutions. Prior experience in finance or markets is not required. Many of us joined without it; we hire for ability, curiosity, and growth trajectory, and we will teach you the markets. This is an entry-level role, and we welcome candidates whose qualifications align with the position. How we work We work in an AI-native way. Directing several coding agents in parallel — to run experiment sweeps, refactor subsystems, or investigate issues across a large codebase — is a normal part of the workflow, and we expect everyone to develop fluency in it. We hold ourselves to a high standard of scientific rigor. Because our results translate directly into capital at risk, sound research methodology is essential: forming clear, testable hypotheses; designing controlled and reproducible experiments; comparing approaches on a consistent, like-for-like basis; and carefully distinguishing genuine signal from noise or overfitting. We value researchers who are skeptical of their own results and who validate their conclusions before acting on them. The systems we work with are complex, and understanding them well is part of the job. Clear communication matters: we collaborate continuously with quantitative researchers and traders across a large organization. Responsibilities: Model research. Design and train the deep-learning models that predict where prices are going — developing new architectures, improving features, and extending how far into the future the models can reliably predict. Enabling ML across the firm. Build the systems and tooling that allow many other researchers to apply machine learning to their own problems, multiplying your impact well beyond your own work. High-impact markets. Take on novel problems in markets that generate billions of dollars in annual revenue, where established solutions do not yet exist. From prediction to trading decisions. A prediction is only useful once it informs a decision to buy or sell. We apply AI across the full chain — including simulating how a set of trades would perform before any capital is committed, and a growing range of decision problems beyond prediction itself. What makes it challenging The problems are difficult and genuinely adversarial: predicting price movements from the fine-grained patterns in how millions of orders arrive and change, extending those predictions further into the future, simulating trading outcomes accurately, and building infrastructure that supports many distinct trading strategies at once. Unlike most AI work, which is evaluated on benchmarks, our work is evaluated by the market — a demanding, high-stakes environment with immediate feedback. The quality of your work is reflected directly and quickly in measurable financial outcomes. Qualifications: Have strong software engineering fundamentals and write clean, efficient, correct code. Have solid depth in machine learning and deep learning, with hands‑on experience training and debugging models. Are results-oriented, with a bias toward action, flexibility, and efficiency. Are genuinely curious about how markets work and why trades are profitable, even without a prior finance background. Communicate clearly and collaborate effectively across a large, technical organization (this role involves substantial collaboration). Are coachable and respond well to feedback; we value growth trajectory over starting point. Are comfortable working in an AI-native way, directing multiple agents to carry out research and engineering, or are eager to learn. Strong candidates may also have experience with (None of the following is required, but any of it is a real plus.) Building applications with LLM agents (e.g., Claude Code, Cursor, or custom frameworks) at meaningful scale. Transformer architectures and modern foundation models; post-training techniques such as supervised fine-tuning, reinforcement learning, or preference optimization. High-performance or large-scale ML systems; GPUs/TPUs; JAX or PyTorch. Large-scale data pipelines and distributed training. A demonstrated record of excellence in a demanding area — for example, competitive programming, mathematics or physics olympiads, Kaggle, significant open-source contributions, or research publications. Basics Education: Bachelor’s, Master’s, or PhD — or an equivalent combination of education, training, and experience — in a relevant field (computer science, mathematics, physics, electrical engineering, statistics, or similar). #J-18808-Ljbffr Citadel Securities