Overview

As a Data Scientist III, you will own the analytical strategy for a core product area and take it all the way to a governed, monitored production system at scale, not just a working notebook.

Responsibilities

• Own the end-to-end analytical framework for a product area, i.e., from problem framing through production deployment and monitoring, with minimal oversight
• Apply LLMs, prompt engineering, and retrieval-augmented generation (RAG) to real product problems — e.g., translating complex targeting/attribution outputs into plain language insights, or building AI-assisted workflows for internal or client-facing tools.
• Evaluate when generative AI is the right tool versus when a simpler model or heuristic will outperform it. This role is expected to have informed judgment here, not default-to-LLM instincts
• Stay hands-on: personally build, code, and develop AI/ML solutions rather than only architecting and delegating
• Build models that scale to production data volumes and real-time serving loads, not just prototypes that work on a sample
• Define and optimize data pipelines, ETLs, and AI model-serving workflows for your domain; this is not handed off to a separate MLOps function
• Implement model monitoring and governance (drift detection, performance tracking, retraining triggers, audit trails) so that production models stay reliable and defensible under compliance scrutiny
• Design measurement approaches for problems where clean ground truth doesn't exist
• Mentor and provide technical guidance to Data Scientist I/II team members; review and approve their modeling and code before it ships.
• Design and defend experimentation approaches (A/B testing, causal inference, uplift modeling) that hold up under stakeholder and compliance scrutiny
• Drive adoption of better MLOps tooling, scalability practices, or modeling methods across the data science team
• Proactively identify whitespace opportunities in product roadmap and bring data-backed recommendations to stakeholders, not just answers to their questions.

Qualifications

• Track record of mentoring or reviewing the work of junior data scientists is a must have
• A strong knowledge of SQL, data structures, and query optimization at a level where you can review others' queries, not just write your own
• Familiarity with AI agent orchestration and multi-step agentic workflows is a plus
• Prior experience in advertising technology, programmatic media buying, or another high-scrutiny/regulated data domain is highly advantageous and good to have.
• A strategic thinker who sets technical direction for a problem area, not just aligns to one
• Comfortable operating in ambiguity where the "right" ground truth doesn't exist yet
• Excellent communication skills, with a track record of influencing product or technical decisions using data
• Genuine mentorship instinct; has grown at least one junior team member's skills, not just supervised their output
• Ownership mentality that extends past "the model works" to "the model is scalable, monitored, governed, and still trustworthy six months from now."
• Doesn't let a research rabbit hole delay a hard deadline

Education

• At least 5 years of hands-on experience in data science and machine learning (both) is a must have
• Hands-on MLOps experience is a must have. You should be comfortable owning a model from training through deployment, serving, and production monitoring, not handing it off once it works in a notebook. This includes experience with model-serving infrastructure, pipeline/ETL orchestration, and monitoring or governance tooling (drift detection, performance dashboards, retraining triggers)
• Proven experience building and scaling machine learning models to handle production data volumes, throughput, and latency requirements is a must have. This means designing for scalability up front, not retrofitting it after something breaks in production
• Hands-on experience with LLMs and RAG architectures is a must have. This includes practical experience with prompt engineering, embedding-based retrieval, and vector databases (e.g., Pinecone, FAISS, Weaviate), and frameworks like LangChain/LangGraph or AWS Bedrock
• Demonstrated ownership of at least one modeling project taken fully end-to-end into production, with measurable business impact