Deep Learning Researcher

Citadel Securities; Miami (FL); On-site; USD 200,000 - 260,000; Full time

Job summary

Citadel Securities in Miami is seeking an AI research scientist to advance deep learning systems used in production trading environments. You will train, fine-tune, evaluate, and scale models to improve performance, reliability, and inference speed for real-time market use cases.

We value researchers who can own problems end-to-end, reason from first principles, and collaborate with engineers and traders to translate insights into measurable market improvements.

Qualifications

• Strong background in machine learning and statistical modeling.
• Experience turning research into production-grade ML systems.
• Proficiency with Python and ML frameworks (PyTorch, JAX).

Responsibilities

• Train, fine-tune, evaluate, and scale models for production-grade market use cases.
• Identify high-frequency signals and improve model performance, reliability and speed.
• Collaborate with cross-functional teams to translate insights into measurable market improvements.
• Design datasets, training flows, and evaluation frameworks for robust deployment.

Skills

Applied ML reasoning

Training dynamics

Statistics & probabilistic modeling

Tools

Python

PyTorch

JAX

Job description

Citadel Securities is a global market-making and liquidity-provision firm, playing a foundational role in ensuring market functionality. The firm’s culture and value proposition are distinct, especially when contrasted with typical frontier AI or Big Tech research organizations:

• Direct Business and PnL Impact: Your work goes directly into production, and its success is measured by the tangible value it generates.
• End-to-End Ownership: Researchers are expected to own problems from inception to production. This includes researching model types and strategic choices, monitoring performance, and iterating improvements.
• Agile, Autonomous Teams: Researchers work in small, nimble teams and are empowered to identify and drive high-value directions into production. This structure eliminates bureaucracy and fosters independent, collaborative exploration.
• Strategic Autonomy and Diverse Exploration: While the team works in close collaboration, a successful researcher is one who can independently identify high-value directions and drive them into production.

This position is focused on advanced deep learning and artificial intelligence research applied to complex market systems, rather than traditional quantitative research. The primary responsibilities center on training, fine-tuning, evaluating, and scaling models to improve performance, reliability, alignment, and model inference speed for production-grade market use cases.

Responsibilities

• Develop and analyze statistical and machine learning models to identify high-frequency signals across global markets.
• Apply mathematical and statistical first principles to creatively adapt models for noisy, low-signal environments where standard techniques are ineffective.
• Design efficient algorithms and dynamic programming solutions to handle high-throughput data. Diagnose training issues such as instability or overfitting, optimizing systems for stability and generalizability.
• Leverage pre-training and fine-tuning methodologies to adapt models to domain-specific datasets. Collaborate with cross-functional teams of researchers, engineers, and strategists to translate technical insights into measurable market improvements. Shape the design of datasets, training flows, and overall model evaluation frameworks.

The interview process is fundamentally designed to evaluate a candidate’s ability to take an unfamiliar problem, break it down from first principles, and debug it independently.

We are implicitly hiring for the ability to reason through ambiguity and off-script scenarios. Candidates must demonstrate a precise understanding of training dynamics, scaling behavior, and distributed failures through live problem-solving.

We evaluate candidates on their ability to solve open-ended, unfamiliar problems (e.g., unexpected training failures, scaling bottlenecks) by building solutions from the ground up. Strong candidates reason from first principles rather than relying on intuition or trial-and-error.

Technical Domain Core Expectations & Focus Areas

• Applied ML Reasoning Ability to analyze unfamiliar machine learning problems systematically across the
• Training Dynamics Solid understanding of model behavior during training, including instability, overfitting, generalization, and bridging the gap between offline validation and production environments.
• Statistics & Probabilistic Modeling
• Practical Implementation Proficiency in writing clean, modular, readable, and reproducible code in Python using PyTorch or JAX.