Senior Data Scientist

The Senior Data Scientist will lead the maturation of Securly's content classification system — building the ML infrastructure that determines, at scale, whether web content is appropriate for K-12 students, and establishing the rigorous evaluation framework that product and leadership teams depend on.

This is applied ML with direct student safety impact — not research. You will lead a significant uplift of Securly's classification models: refactoring binary models to proper multiclass classification, building labeled evaluation datasets, and producing standardized model cards with per-category precision, recall, F1, and confusion matrix analysis.

At L5, you are the technical leader of the data science function for content safety. You will define the evaluation methodology the team follows, set the standard for what a model card must contain before a model ships, mentor the team on applied ML rigor, and serve as the interface between data science and engineering on production integration constraints.

L5 at Securly is a Staff Engineer. You are the technical owner, not just an implementer.

• Drive technical direction for your initiative end-to-end: from architecture to production, with minimal oversight from your engineering manager.
• Identify and resolve ambiguity in requirements, system boundaries, and design tradeoffs without waiting for a fully-formed spec.
• Mentor L3/L4 engineers on the team: code reviews, design feedback, pairing, and raising the bar for what production-quality work looks like.
• Partner with your L6 technical lead and the Distinguished Engineer on architectural decisions, surfacing tradeoffs clearly rather than deferring them upward.
• Contribute to cross-team engineering standards: you are expected to influence practices beyond your immediate squad.
• Translate technical context into clear written artifacts that non-engineers (PM, Support, Leadership) can act on.
• Participate in on-call rotation and own the full incident lifecycle for your system: detection, diagnosis, resolution, and retrospective.

• Define the evaluation methodology for content classification at Securly: establish what a model card must contain and hold every model release to that standard before it ships.
• Lead the multiclass refactor of Securly's content classification models: redesign binary models to handle multi-label, multi-class content categories (Adult Content, Violence, Self-Harm, Social Media, and others).
• Build and maintain labeled evaluation datasets with robust annotation workflows; address class imbalance and label noise systematically; document dataset curation decisions in a versioned data card.
• Connect offline evaluation to production monitoring — surface classification drift and error patterns before they become customer-facing issues.
• Investigate and resolve misclassification errors: false positives (over-blocking) and false negatives (under-blocking); produce written root cause analyses.
• Build and maintain training data pipelines: ingestion, cleaning, labeling, and versioning at scale.
• Mentor the existing AI team on evaluation methodology, model development practices, and data science communication rigor.
• Communicate precision/recall tradeoffs to product managers and engineers; produce executive-level summaries of classification quality for leadership.
• Collaborate with engineering to integrate model outputs into the production filtering stack with appropriate latency and reliability constraints.
• Research and prototype improvements: feature representations, model architectures, active learning for label efficiency, domain adaptation for emerging content categories.

• You have shipped ML models to production and lived with the consequences — you know what model drift looks like and how to catch it before it becomes a customer issue.
• You treat evaluation as a first-class engineering artifact. A model without a model card is not finished — and you set and enforce that standard for the team.
• You define the methodology, not just apply it. You produce the evaluation framework that other data scientists use, and you hold them to it.
• You can communicate precision/recall tradeoffs to a product manager and to a senior engineer in the same conversation, calibrated to each audience.
• You are energized by problems with real stakes: a false negative in Self-Harm classification is not an acceptable error rate.
• You mentor by example and by expectation: your code, your analysis, and your documentation set the standard.