Senior Applied Scientist - Behavior AI

The Behavior AI team builds the AI-based anomaly detection behind Datadog's security products. Our models learn what normal looks like across the billions of logs, events, and telemetry records flowing through the platform every second, and they flag the behavior that does not fit, on every record, in real time, at a cost that makes sense at our scale. What we build does not ship to a single feature. The same models power detection across many of Datadog's security products at once, so the work has impact well beyond any one team. Large general-purpose models are too slow and too expensive to run in that path, so we take the opposite approach: small, custom models, designed for high-throughput stream processing and optimized to run cheaply on every record. We are hiring a Senior Applied Scientist to build these models from start to finish. You will contribute to designing the architecture, training at scale, and the optimization work that takes a model from training to running efficiently on production traffic. This optimization requires a deep understanding of the constraints imposed by both the software and the hardware, together with the applied mathematics to work within them: often it comes down to finding a mathematical reformulation that fits those constraints better, and that judgment can decide whether a model reaches production at all. The work involves a number of open questions. How do you obtain most of the quality of a large model from one that is far smaller and cheap enough to run on the full stream? Where is it worth trading exactness for speed, and how do you reason about the error you accept? How do you make a small model's outputs clear enough that the detection engineers and analysts who rely on it can trust what it reports? If these are the problems you want to work on, we would like to hear from you. At Datadog, we place value in our office culture: the relationships and collaboration it builds, and the creativity it brings to the table. We operate as a hybrid workplace to ensure our Datadogs can create a work-life harmony that best fits them. What You'll Do: - Design and build custom mid-size models for high-throughput stream processing, and train them at scale. - Optimize these models from start to finish, working across both the mathematics and the systems, often by finding mathematical reformulations that fit the hardware and software constraints better. - Build the training data pipelines the work depends on when they do not already exist, from the raw stream to a training-ready dataset. - Work with engineering to integrate models into production, with a clear focus on GPU utilization, latency, and cost per record on live traffic. - Plan the roadmap of model and system improvements, based on a solid understanding of the product and of what matters most to users. - Build an agentic layer on top of the models to analyze, validate, and act on their outputs. - Build lightweight interpretability tools that make model behavior easier to explain to the people who rely on it. - Maintain and monitor the models, services, and infrastructure your team owns, and take part in your team's on-call rotation. Who You Are: - You have a BS/MS/PhD in Computer Science, Engineering, Machine Learning, Applied Mathematics, or a related scientific field, or equivalent experience. - You have hands-on experience training and fine-tuning models at scale, and deploying them into production systems with real throughput and cost constraints. - You have a working knowledge of how GPUs work, and a track record of making models run efficiently within real hardware constraints. - You have strong applied-mathematics fundamentals and reach for them naturally when designing and optimizing models. - You have a real passion for applied mathematics, software design, and implementation. This role sits at the intersection of the three, and it is a requirement for the position. - You care about code simplicity and performance, and you can build the data pipelines and the surrounding production code, in addition to the models themselves. - You can explain complex ideas and trade-offs clearly to engineers and product partners, and you let a solid understanding of the product guide what you build next. - Bonus: experience with efficient sequence architectures, model interpretability, or large-scale streaming systems. Datadog values people from all walks of life. We understand not everyone will meet all the above qualifications on day one. That's okay. If you're passionate about technology and want to grow your skills, we encourage you to apply. Benefits and Growth: - New hire stock equity (RSUs) and employee stock purchase plan (ESPP) - Continuous professional development, product training, and career pathing -

BS/MS/PhD in Computer Science, Engineering, Machine Learning, Applied Mathematics, or a related scientific field, or equivalent experience. Hands-on experience training and fine-tuning models at scale, and deploying them into production systems with real throughput and cost constraints. Working knowledge of how GPUs work, and a track record of making models run efficiently within real hardware constraints. Strong applied-mathematics fundamentals and ability to apply them to design and optimization of models. Passion for applied mathematics, software design, and implementation. Care for code simplicity and performance, with ability to build data pipelines and surrounding production code in addition to models. Ability to explain complex ideas and trade-offs clearly to engineers and product partners, guided by product understanding. Bonus: experience with efficient sequence architectures, model interpretability, or large-scale streaming systems.

Design and build custom mid-size models for high-throughput stream processing, and train them at scale. Optimize these models from start to finish, often by finding mathematical reformulations that fit hardware and software constraints. Build training data pipelines from raw stream to training-ready datasets. Work with engineering to integrate models into production, focusing on GPU utilization, latency, and cost per record on live traffic. Plan the roadmap of model and system improvements based on product understanding and what matters to users. Build an agentic layer on top of the models to analyze, validate, and act on outputs. Build lightweight interpretability tools to help explain model behavior to users. Maintain and monitor models, services, and infrastructure, and participate in on-call rotation.