Data Scientist

What matters most

• Deliver production AI applications

  The posting calls for designing, deploying and supporting enterprise-grade AI services, with Python backends, REST APIs, microservices and cloud-native practices. Prepare a concrete account of taking a solution from concept through deployment and support.

• Build retrieval-grounded LLM solutions

  LLMs, RAG, embeddings, vector databases and semantic retrieval are central to the role. Prepare to explain how you would design retrieval, connect it to generation and assess whether the results answer the business need.

• Orchestrate agents for business processes

  The role includes agent and multi-agent workflows, tool and function calling, and business process automation. Prepare to explain orchestration choices and how agents interact with enterprise systems.

• Make AI safe and measurable

  The posting explicitly calls for guardrails, access controls, human-in-the-loop workflows, evaluation, testing, monitoring and observability. Prepare to describe how these controls and measurements would work in a deployed application.

• Translate requirements into reusable solutions

  Collaboration with product owners, architects and engineers—and contributions to reusable frameworks and platform capabilities—are stated duties. Prepare an example of turning a business requirement into a scalable design while keeping components reusable.

Likely challenges to think through

• Retrieval quality versus operating cost

  Preparation hypothesis, not a stated Wesco problem: improving RAG answer quality may require more retrieval and context, increasing latency and cost. Be ready to describe how you would evaluate relevance and answers, then tune the design against those trade-offs.

• Agent autonomy versus control

  Preparation hypothesis: agents that can call tools and automate workflows create a tension between flexibility and security. Prepare to explain where you would enforce access controls, limit actions and require human review, including for prompt-injection risks.

• Rapid iteration versus production reliability

  Preparation hypothesis: changing prompts, models and agent workflows can make behavior harder to test and support. Prepare to explain how evaluation, CI/CD, observability and ongoing monitoring would help detect regressions without halting useful iteration.

Topics to prepare

• LLM behavior and prompt engineering

  Revise how instructions, context and prompt structure affect outputs; compare two prompts on the same task and explain the difference. The posting requires LLM and prompt-engineering understanding and calls for improving prompt effectiveness.

• RAG and semantic retrieval

  Sketch how documents become embeddings, how a vector database retrieves relevant content and how that content reaches an LLM. Explain likely retrieval failure points. RAG, embeddings, vector databases and semantic retrieval are explicit requirements.

• Agent design and orchestration

  Diagram a business-process workflow with tool calling, agent memory and a handoff between agents; explain when a simpler workflow would suffice. The role calls for agent design patterns, multi-agent orchestration and function calling.

• AI security and governance

  Walk through a prompt-injection attempt against an agent using enterprise data, identifying access controls, guardrails and a human approval point. These controls and responsible AI are explicitly named in the posting.

• Evaluation and operational reliability

  Define how you would assess answer quality, retrieval quality, prompt effectiveness, user experience, monitoring signals and cost for an AI service. The posting calls for evaluation, observability, hallucination mitigation and cost optimisation.

• Production AI architecture

  Draw an architecture connecting a backend service, an LLM service, search, enterprise APIs and a cloud deployment; explain scaling and failure handling. The role emphasizes production-grade applications, integration and cloud-native design.

Skills to revise

• Build a Python AI API

  Practice implementing a small REST endpoint in Python with FastAPI, Flask or a similar framework, including input validation and tests. Strong Python, backend-framework experience, REST APIs and testing are stated requirements.

• Implement and inspect a RAG flow

  As a revision exercise, index sample documents, retrieve passages through embeddings and a vector database, and inspect whether generated answers use the retrieved context. The posting explicitly calls for building LLM, RAG and semantic-retrieval solutions.

• Prototype a controlled agent workflow

  Implement a small workflow that calls a mock business API, passes context between steps and pauses for human approval before a consequential action. Tool calling, enterprise integration, orchestration and human-in-the-loop controls are all in scope.

• Test and observe an AI service

  Create a compact set of expected responses and failure cases, then record outcomes and useful monitoring signals after changing a prompt or retrieval setting. The role calls for AI evaluation frameworks, testing, monitoring and observability.

• Rehearse a production delivery path

  Use Git and a CI/CD pipeline to run tests and prepare a small service for deployment on one of the listed cloud platforms. This is practice for the posting’s cloud-native deployment, CI/CD and production-support expectations; no particular cloud provider is mandatory.

• Map evidence to the stated requirements

  Create a one-page inventory of applicable examples covering software engineering, hands-on generative AI, production support, Python APIs, cloud deployment, testing and CI/CD. The posting specifies 5–8 years of software engineering and at least 2 years of generative AI experience; do not claim experience you cannot substantiate.

• Prepare an end-to-end project walkthrough

  If you have a relevant project, diagram its requirements, LLM or RAG architecture, your contributions, deployment and operational results. Be ready to explain trade-offs and what changed after testing or user feedback.

• Practice an enterprise RAG design exercise

  Sketch how an application could connect enterprise data to embeddings, semantic retrieval, a vector database and an LLM through a Python REST service. Explain access controls, cloud and search-platform choices, and how you would validate the design; this is a preparation exercise, not a stated Wesco project.

• Design a controlled agent workflow

  Draw a proposed tool-calling or multi-agent flow for business process automation. Identify where context or memory is needed, how to limit prompt-injection risks, and when a human should review or approve an action.

• Build an evaluation and operations plan

  Prepare a sample test set and explain how you would assess retrieval, answer quality, hallucinations, prompts and user experience. Add monitoring, observability, security, cost and CI/CD checks for a production-grade service.

• Rehearse collaboration and reuse examples

  Prepare a concise example, if available, of translating a business requirement with product owners, architects or engineers into a scalable solution. Identify any reusable framework or platform component and explain its value without overstating your role.

• Write questions for the employer

  Bring questions about the AI applications this role would support, available enterprise data and APIs, the cloud and search stack, evaluation standards, human-in-the-loop controls, and ownership after deployment.