Sr Data Scientist - AI Engineer

What matters most

• Deliver decision-ready analytical solutions

  Solution development is the largest responsibility (35%): design models for large datasets, select and interpret methods, and communicate recommendations and their impact. Prepare to explain how an analysis led to a merchandising decision.

• Build production-grade agentic AI

  The role is tasked with establishing MLOps, LLMOps and AIOps foundations and moving agents from experiments into governed production workflows. Prepare to explain an architecture that connects enterprise data, retrieval, models, tools and APIs with testing, monitoring and rollback.

• Direct project work without direct reports

  Project management and team support account for 30% of the role: define goals with partners, prioritize and assure work quality, help allocate resources, mentor junior colleagues and support hiring. Prepare an example of guiding delivery through collaboration rather than formal reporting authority.

• Connect technical work to merchandising needs

  Business collaboration accounts for 20%: apply business knowledge, build cross-functional trust, educate partners on analytics and identify opportunities. Prepare to translate a technical recommendation into a business decision and explain its limits to a non-technical audience.

• Make solutions reusable

  Technical exploration and development account for 15%: assess new techniques and data sources, document code and processes, and establish practices for analysis and model production. Prepare to discuss how a prototype became a documented, reusable component.

Likely challenges to think through

• Agent autonomy versus governance

  Preparation hypothesis: Multi-step agents that call tools and use retrieval may be useful but harder to trace and control. Be ready to discuss how evaluations, tool-use auditing, guardrails, human approval and fallback mechanisms could balance capability with reliability; these practices appear in the preferred qualifications.

• Speed of experimentation versus production rigor

  Preparation hypothesis: Moving an AI prototype into a supported workflow may require trading rapid iteration for automated tests, versioning, observability and rollback. Prepare a concrete account of release decisions, failure detection and coordination with engineering or operations partners.

• Competing demands on a senior individual contributor

  Preparation hypothesis: Hands-on solution development could compete with prioritization, stakeholder communication and mentoring, all listed duties despite the role having no direct reports. Prepare to explain how you would make trade-offs visible, protect work quality and delegate or sequence work with the team.

Topics to prepare

• Agent architecture and orchestration

  The posting prefers multi-step agents with tool calling, planning, state, memory, structured outputs and API integration. Sketch an agent that answers a merchandising question using a model and an API; mark where state is retained and where a human must approve an action.

• Retrieval and enterprise data

  The role combines enterprise data, retrieval systems and embeddings. Compare when an agent should retrieve source material versus call a data-science model, then diagram how it would cite or record the evidence behind a recommendation.

• Production AI lifecycle

  The posting calls for MLOps, LLMOps and AIOps foundations, including versioning, registries, CI/CD, retraining and rollback. Trace a model or agent change from evaluation through release and rollback, identifying what must be recorded at each step.

• Observability and governance

  The preferred qualifications name tracing, alerting, evaluation, safety guardrails, tool-use auditing, fallback mechanisms and cost/latency monitoring. Work through a failed agent-tool call: decide what a trace should show, when to alert, and when to fall back or require human review.

• Merchandising analytics and business impact

  The role designs models for large datasets and turns findings into recommendations; merchandising or related domain experience is preferred. As a preparation exercise, frame a hypothetical assortment decision, select an analytical approach, and state how its result would inform a business choice without assuming access to company data.

Skills to revise

• Python and SQL analysis

  Both are named in the preferred software-engineering skills. Using a small synthetic merchandising dataset, write SQL to prepare features and Python to fit or evaluate a model; summarize the result and its limitations.

• Agent prototype and tool integration

  The posting prefers hands-on agent workflows, API integration and lightweight tool prototypes. Build a small Python prototype that calls a mock tool, handles a failed call, and returns a structured result; document the tool contract and state transitions.

• Automated agent evaluation

  Automated testing and evaluation harnesses are explicitly listed. Create test cases for correct tool selection, unsupported answers and malformed tool responses; record pass criteria and inspect failures rather than relying on a single successful demonstration.

• Deployment and recovery practice

  Docker, Kubernetes, cloud deployment, CI/CD and rollback are preferred capabilities, with GCP preferred among cloud options. Containerize a small service and rehearse a versioned release and rollback; this is a practice exercise, not a claim about The Home Depot’s deployment setup.

• Technical communication and reusable documentation

  The responsibilities include presenting recommendations to technical and non-technical partners and documenting reusable solutions. Prepare a one-page design note and a brief business-facing explanation for the prototype, covering the decision it supports, evaluation evidence, operational risks and next steps.

• Build a role-to-evidence sheet

  The posting specifies five years of work experience and prefers production AI and hands-on GenAI or agentic AI experience. Prepare a one-page map of its stated requirements and preferences to examples you can discuss; leave gaps explicit rather than inventing evidence.

• Rehearse a production AI walkthrough

  The role focuses on moving agentic solutions from experiments into production. Prepare a project walkthrough, if you have one, covering the problem, your contribution, tools and retrieval, evaluation, deployment, and measured results. Otherwise, label your example as a proposed design.

• Sketch a merchandising agent design

  As a preparation exercise, diagram how an agent could reason over enterprise data and orchestrate models, APIs, and retrieval for a merchandising decision. Explain tool permissions, state, structured outputs, failure handling, and where a person would approve an action; do not present the design as Home Depot’s architecture.

• Prepare an operations and governance plan

  The posting calls out MLOps, LLMOps, AIOps, observability, and governance. Draft a release checklist covering tests and evaluations, versioning, monitoring, alerts, tool-use auditing, safety controls, rollback, and production support, with tradeoffs you could explain.

• Practice turning analysis into a recommendation

  Solution development includes choosing methods for large datasets and communicating business impact. Prepare a short presentation showing a method choice, key finding, limitations, recommendation, and how you would explain it to both technical and non-technical partners.

• Gather collaboration and mentoring examples

  The role includes prioritizing work with business partners, assuring quality, mentoring junior colleagues, and supporting hiring, but has no direct reports. Prepare concise examples of how you coordinated decisions, improved another person’s work, or documented a reusable solution.

• Write questions for the employer

  Prepare questions about the team’s current path from prototype to production, priority merchandising decisions, evaluation and approval practices, engineering partnerships, and how success is measured. These are questions to ask, not claims about existing processes.