Principal Data Scientist

We are hiring a Principal Data Scientist for the Data Science group, reporting directly to the VP Engineering. This is a senior individual-contributor leadership role: you own the technical direction and architecture of what the group ships - real-time inference at scale, automated red-teaming of GenAI systems, and the Databricks/Spark data platform underneath - and you lead the hardest projects end to end, hands on in the code.

What you'll own

A large Python, Rust, and PySpark monorepo, dozens of production services and shared libraries, across three connected domains:

Content-moderation inference at scale

Real-time, multi-tenant detection across text, image, video, and audio - hate speech, CSAM, nudity, child grooming, extremism, PII, prompt injection, age estimation, and more - served through our in-house ActiveServe framework over NVIDIA Triton and a Rust detection service, on latency-sensitive, SLA-bound, customer-facing traffic with per-customer custom models.

GenAI safety and red-teaming

Automated red-teaming that attacks customers' LLM applications with a research-driven attack taxonomy and measures attack-success rate, alongside the defensive side - LLM-as-judge escalation to cut false positives, and the tooling that authors and refines moderation policies - built on a multi-provider LLM foundation (Bedrock, Anthropic, OpenAI, Gemini, xAI Grok).

Data platform and MLOps

An end-to-end lakehouse and MLOps stack on Databricks - bronze/silver/gold ingestion, PySpark pipelines, and model training, versioning, and promotion through MLflow / Unity Catalog into Triton serving - with the performance-critical hot paths engineered in Rust (PyO3/maturin) for sub-millisecond, high-QPS matching and detection.

What you'll do

• Own the GenAI safety roadmap - decide what we build with LLMs and what we build around them: guardrails and LLM-as-judge adjudication, prompt-injection and jailbreak detection, policy authoring, and the automated red-teaming that measures attack-success rate against customers' applications. Choose which model families and providers we use, fine-tune, or retire, and what each has to prove before it reaches customer traffic.
• Own how we evaluate GenAI systems - benchmark and dataset design over real production traffic, LLM-jury labeling protocols, multilingual, multi-turn, and multimodal coverage, calibration and threshold selection, and the error analysis that turns a metric into a fix. Run the evaluations that separate a real shift in what these models can do from a launch post, and keep quality honest in production as traffic, prompts, and attacks drift.
• Lead the hardest GenAI work end to end, and grow the craft - take the ambiguous, high-stakes projects from framing through prototype, offline evaluation, shadow traffic, and rollout; mentor data scientists and ML engineers through design, experiment, and model review; and translate model behavior into what Product and customers care about: accuracy, latency, cost, and enterprise readiness.

What we're looking for

• ML technical leadership - a proven tech lead, senior data scientist, or ML engineer who has taken multi-person, multi-quarter ML initiatives from framing to production without formal authority, with strong Python and the written communication to make the case in a design or an evaluation.
• Deep understanding of LLM technology - how these models are trained, adapted, and served, what prompting, fine-tuning, RAG, and agentic patterns each buy you, where they fail, and how to measure quality, latency, and cost.
• AI-augmented engineering - deep, daily fluency with an AI coding assistant (Claude Code, Cursor, or Codex), with the judgment to raise the whole team's leverage and set how these tools are used well.