AI Developer

What matters most

• Ship useful AI applications and agents

  The role designs and deploys LLM applications and autonomous or multi-agent workflows, including tool use, RAG, and multi-step reasoning. Prepare to explain design choices that take an agent from a prototype to a working application.

• Ground AI in trusted data

  The role builds pipelines connecting Snowflake and Azure Databricks to AI/ML workflows so models and agents use well-structured data. Prepare to discuss data quality, transformations, and retrieval choices; dbt and unstructured-data processing are preferred experience, not required.

• Measure quality before and after launch

  The posting calls for evaluation harnesses and tests covering model and agent quality, reliability, and safety before and after deployment. Prepare a concrete approach to test cases, failure analysis, and ongoing evaluation.

• Build maintainable production software

  Required qualifications emphasize tested Python code, SQL, a cloud data platform, and software fundamentals such as version control and CI/CD. Prepare to describe how these practices support deployment and maintenance rather than just a successful demo.

• Turn uncertain needs into business value

  As a foundational member of a new AI team, the developer will work with data science, engineering, IT, and business stakeholders, prototype emerging tools, and recommend adoption where value is clear. Prepare to explain how you would narrow an ambiguous request into a testable use case.

Likely challenges to think through

• Balancing fast prototypes with production rigor

  Likely challenge, not a confirmed employer problem: rapid tool evaluation could compete with testing and maintainability. Prepare a decision framework for when to continue a prototype, harden it for deployment, or stop it.

• Containing agent failures without losing utility

  Likely challenge: tool-using, multi-step agents can accomplish more than simple LLM calls but introduce more failure paths. Prepare to discuss bounded tool permissions, checks on retrieved information, and evaluations that expose unsafe or unreliable behavior.

• Coordinating data access across platforms

  Likely challenge: connecting Snowflake, Azure Databricks, and AI workflows while supporting privacy, access control, and governance may create tension between useful data access and appropriate restrictions. Prepare to explain how you would identify data owners, validate permitted access, and test what an application can retrieve.

Topics to prepare

• Agent orchestration and tool use

  Posting: The role develops autonomous and multi-agent workflows with Claude Agent SDK and Azure AI Foundry Agent Service. Revise by diagramming an agent that selects a tool, handles a failed call, and completes a multi-step request; explain where each framework would fit.

• Retrieval-augmented generation

  Posting: Agent workflows include RAG; embeddings, chunking, semantic search, and structuring unstructured data are preferred experience. Revise by comparing two chunking approaches for a sample document and explaining how retrieval results would be checked before an answer is generated.

• Trusted data for AI workflows

  Posting: Pipelines connect Snowflake and Azure Databricks to AI/ML workflows and must provide trusted, structured data. Revise by sketching the data flow, identifying transformations and validation points, and explaining how an agent would use the resulting data.

• Model and agent evaluation

  Posting: Evaluation must measure quality, reliability, and safety before and after deployment. Revise by defining a small test set, expected outcomes, failure categories, and a way to compare results across versions.

• Privacy, access control, and governance

  Posting: The role supports responsible AI within Azure and Snowflake environments. Revise by tracing how a restricted data field could reach an agent response and identifying access checks and tests that would prevent disclosure.

Skills to revise

• Build a tested Python LLM application

  Required: Strong Python, maintainable tested code, and hands-on exposure to generative AI APIs. Practice implementing a small Python application that calls Claude or Azure OpenAI, separates API handling from application logic, and tests success and error paths.

• Write SQL for an AI data feed

  Required: Working SQL knowledge and experience with a cloud data platform. Practice writing queries that join, filter, and validate sample records for an AI workflow; explain how the result could feed a Snowflake–Azure Databricks pipeline. dbt is mentioned as a pipeline tool and preferred experience, not a requirement.

• Prototype a tool-using agent

  Posting: The role builds agentic workflows with tool use and multi-step reasoning. Practice a small prototype using the Claude Agent SDK or Azure AI Foundry Agent Service, with one constrained tool and a defined response when the tool fails. Direct experience with these frameworks is preferred.

• Implement an evaluation harness

  Posting: The role builds testing frameworks for model and agent behavior. Practice running a fixed set of prompts through a prototype, recording outputs and pass/fail reasons, and rerunning the set after a change to detect quality or safety regressions.

• Exercise production software practices

  Required: Familiarity with Git/GitHub, testing, CI/CD, and design patterns. Practice making a small change on a branch, adding an automated test, and explaining how a CI/CD check would catch a regression before deployment.

• Turn an ambiguous request into a prototype plan

  Posting: The role works with data science, data engineering, IT, and business stakeholders and evaluates prototypes for business value. As interview preparation, take a hypothetical AI workflow request, list clarifying questions, define a narrow prototype and measurable outcome, and identify what each stakeholder would need to confirm.

• Map your evidence to the requirements

  Prepare a one-page index of real examples covering production delivery, tested Python, SQL and a cloud data platform, LLM API use, Git, CI/CD, and learning a new tool. Note where each example shows your own contribution.

• Rehearse a production project walkthrough

  Select one software, data, or AI/ML application you shipped. Be ready to explain the need, architecture, your decisions, testing and deployment, results, and what you would change.

• Sketch an agent workflow

  As a practice exercise, draw a workflow using an LLM, tool calls, retrieval, and multi-step reasoning. Explain when you would use a single agent versus multiple agents, and how you would handle failed or unauthorized tool calls.

• Trace the data from source to answer

  Prepare a small design showing how structured data could move between Snowflake, Azure Databricks, and an AI workflow. Include SQL transformations, data-quality checks, and, where relevant, how unstructured documents would be chunked and retrieved. Treat this as a proposed design, not Diamondback’s existing architecture.

• Draft an evaluation and safety plan

  Define a small test set and measures for answer quality, reliability, and safety before and after deployment. Show how you would investigate failures and address data privacy, access control, and governance in Azure and Snowflake.

• Prepare a prototype decision story

  Use a real example, if available, of learning and comparing unfamiliar tools. Explain the business need, what you tested, the trade-offs you found, and how you communicated a recommendation to technical and business stakeholders.

• Write questions for the new AI team

  Bring questions about its first use cases, available trusted data, how stakeholders define success, and who owns deployment, evaluation, and access decisions. These clarify the role’s stated ambiguity without assuming established processes.