Senior ML Engineer

Role overview

A senior engineering role focused on building the safety models that sit on the hot path of an AI product. These models interpret what an AI system is attempting to do and choose whether to permit, modify, or stop the action — spotting jailbreak attempts, catching sensitive data before it leaks, and separating routine calls from genuinely risky ones. The position owns the work end to end, from framing the underlying question through to a model running reliably under live customer load, where latency and correctness both matter and mistakes are visible.

Responsibilities

- Drive detection models from initial problem framing through data work, training, evaluation, and production deployment. - Tune models for tight latency budgets so safety checks do not become a bottleneck on user traffic. - Build and refresh evaluation suites and red-team datasets that keep model quality honest as adversarial inputs change. - Partner with engineering and product teams so models integrate cleanly into the surrounding system. - Operate models in production: monitor for drift, catch regressions, and ship updates safely. - Work with product owners to turn ML trade-offs into clear, confident decisions, including when a model should defer to a human reviewer. - Mentor peers, review work, and translate recent research into practical improvements in the product.

Requirements

- Master's or PhD in Computer Science, Data Science, Mathematics, or a closely related field, or a Bachelor's degree combined with 8+ years of directly relevant industry experience. - At least 5 years building and shipping ML systems in production, with genuine ownership of models that served live traffic. - Applied background in NLP or LLM safety — classification, sequence modeling, or large-scale content moderation. - Deep Python skills and strong command of the modern ML stack (PyTorch, Hugging Face Transformers, scikit-learn, spaCy), including comfort going below the API surface. - Hands-on experience with large-scale data tooling such as Spark or PySpark, Ray, Pandas, Apache Arrow, or Kafka to build the pipelines feeding training and inference. - Experience serving models at low latency using systems like vLLM, ONNX Runtime, or NVIDIA Triton, including quantization and distillation to meet latency targets. - Fluency with production infrastructure (Docker, Kubernetes, CI/CD, AWS or GCP) and the data stores behind ML systems (PostgreSQL with pgvector, Redis, S3, Snowflake). - Familiarity with experiment tracking and orchestration tools such as Weights & Biases, MLflow, Airflow, or DVC, with disciplined reproducibility practices. - Strong performance instincts — attention to tail latency, not just averages — and a habit of refusing to trust a model that cannot be properly measured. - Clear communication skills for explaining complex ML problems and solution paths to non-technical stakeholders. - Comfort owning open-ended problems on a fast-moving, early-stage team.