Director of Machine Learning, Narvar

• We care more about judgment and ownership than credentials
• Have 12+ years in engineering with 5+ years managing ML or data teams, including managing managers or senior ICs
• Are an engineer at heart — you can still read a training pipeline, review a feature spec, and tell when an eval is measuring the wrong thing
• Have shipped ML systems that make consequential automated decisions in production, and have owned them after launch — drift, retraining, incidents, and all
• Have deep experience with at least one of: entity resolution / identity graphs, fraud and abuse detection, risk scoring, or anomaly detection at scale
• Understand what makes ML different from software: labels are noisy and delayed, systems fail silently, offline metrics lie, and last quarter's model is fighting last quarter's adversary
• Have opinions about evaluation — precision/recall tradeoffs on heavily imbalanced data, holdout hygiene, feedback loops where the model's own decisions contaminate future labels
• Have built or scaled ML infrastructure: feature pipelines, training orchestration, online model serving with real latency budgets, monitoring and alerting
• Are fluent in Python and comfortable in a modern data stack (Spark, Airflow or equivalent, streaming, cloud data warehouses — we run on GCP)
• Have hired well, retained well, and can point to engineers whose careers got better on your watch
• Excel in a fast-paced environment with a flat structure — you set direction by earning trust, not by mandate
• Have a BS/MS in computer science, statistics, or an equivalent background

Desirable

• Own the ML charter across identity resolution, fraud and abuse detection, risk scoring, and consumer intelligence — strategy, roadmap, and delivery
• Build and grow a high-performing, globally distributed team of ML engineers; hire, coach, and develop senior ICs and managers
• Set the technical bar for how models get built, evaluated, deployed, and monitored — and hold the org to it
• Push identity resolution coverage, precision, and profile classification accuracy against a measured, frozen-holdout baseline — not against vibes
• Own IRIS model performance end-to-end: detection rate, false-positive rate, label quality, and the feedback loops that keep both honest as fraud patterns shift
• Build the ML platform layer — feature stores, training pipelines, model registry, online serving, drift and performance monitoring — so model velocity isn't bottlenecked on infrastructure
• Partner with the AI engineering team so identity and risk signals are first-class inputs to NAVI's agent decisions
• Work directly with Product, Engineering, Security, and Customer Success to translate messy retailer problems into ML problems worth solving — and to say no to the ones that aren't
• Own build-vs-buy and data-partner decisions (third-party identity data, enrichment providers), including the economics
• Communicate model performance, risk, and tradeoffs credibly to executives, retailers, and the board