New opportunity

AI Engineer

HAVI · Chicago, United States

About this role

HAVI seeks an AI Engineer to design, build and deploy scalable, production-ready AI solutions for predictive and prescriptive analytics in supply chain operations. The role spans prototyping, production deployment and ongoing operation, supporting work in promotions, demand and supply planning, item- and menu-level analytics, simulations, optimization, competitive benchmarking and root cause analysis.

Responsibilities

Partner with data management, data science, decision science and technology teams to address planning, replenishment, pricing and optimization needs; translate business problems into AI solutions.
Collect, clean and preprocess large structured and unstructured datasets. Design and train machine learning and deep learning models, choosing approaches such as classification, regression, NLP and computer vision as appropriate.
Deploy models through APIs or cloud platforms, monitor performance and retrain when needed. Evaluate emerging research and architectures, including transformers and GANs, and implement agentic AI capabilities. Collaborate across analytics, DevOps, product management and technology to operationalize AI and data products.

Qualifications

Bachelor’s degree in computer science, data science, information systems, information science or a related field. An advanced degree in one of these fields is preferred.
At least 2 years of hands-on Azure Databricks experience using PySpark or Scala, Spark SQL and Delta Lake. Experience with Azure Data Factory orchestration and integration with ADLS Gen2 and Key Vault; strong SQL and large-dataset performance-tuning skills.
Experience with large-scale data quality, monitoring, alerting and debug/backfill processes; Git-based development, code reviews, testing and CI/CD; infrastructure as code using Terraform or Bicep; and observability and cloud cost/performance management. Proven cross-functional collaboration to deliver and support data products is required. Relevant AI, ML, cloud or MLOps certifications are a plus.

Location and compensation:

Hybrid role at 345 N Morgan St, Chicago, IL 60607. Candidates must live in the Chicago metropolitan area; relocation assistance is not offered.
Salary range is $105,000–$125,000 with a 5% targeted bonus. Benefits listed include medical, dental and vision coverage, paid leave, a 401(k) with company match and profit sharing, 15 days of paid time off plus holidays, tuition reimbursement and student loan repayment assistance.

For full details about this role, click “Apply on company site” to view the employer’s original posting.

Skills for this role

Machine LearningDeep LearningData EngineeringAgentic AIClassificationRegressionNatural Language ProcessingComputer VisionTransformersGANsTensorFlowPyTorchScikit-learnPythonPySparkScalaSpark SQLSQLAzure DatabricksDelta LakeAzure Data FactoryADLS Gen2Azure Key VaultGreat ExpectationsDeequMLflowKubeflowAWSAzureGoogle Cloud Platform (GCP) 即

Checking your profile…

Preparation suggestions, not the employer’s actual interview questions. Always check the original posting for current requirements.

Understand the role

What matters most

Translate supply-chain needs into AI solutions
The role works with data management, data science, decision science and technology teams on planning, replenishment, pricing and optimization. Prepare to explain how you would define a business problem, choose a useful output and align stakeholders on what the solution should support.
Build scalable Azure data foundations
The posting calls for hands-on Azure Databricks with PySpark or Scala, Spark SQL and Delta Lake, plus Azure Data Factory integration with ADLS Gen2 and Key Vault. Prepare a concrete account of pipeline design, SQL tuning and handling large structured or unstructured datasets.
Take models from prototype into operation
HAVI describes an analytics lifecycle spanning model design, training, production deployment, monitoring and retraining. Prepare to explain how you would select an approach for a stated problem and determine whether it continues to perform after deployment.
Support reliable, maintainable data products
The qualifications specify data-quality checks, alerting, debug and backfill practices, testing, CI/CD, infrastructure as code, observability and cloud cost control. Prepare to discuss how these practices fit together when a data or model pipeline needs ongoing support.
Evaluate emerging AI where it serves a purpose
The responsibilities include experimenting with architectures such as transformers and GANs and implementing agentic AI capabilities. Prepare to distinguish a promising experiment from an approach worth deploying for a specific supply-chain use case.

Likely challenges to think through

Balancing scale, data quality and compute cost
A likely challenge, not a confirmed HAVI problem, is that large datasets can make validation and fast processing expensive. Prepare to explain trade-offs among Spark joins, partitions, file sizing and cluster configuration while preserving trustworthy outputs.
Crossing the prototype-to-production gap
A likely challenge is reconciling model experimentation with repeatable deployment and ongoing monitoring. Prepare an example of how you would set performance checks, detect degradation and decide when retraining or investigation is warranted.
Choosing the right level of AI complexity
The mix of predictive analytics, optimization and emerging AI suggests a possible trade-off between a newer architecture and a simpler, easier-to-operate method. Prepare to justify a choice using the business objective, available data, evaluation results and operational burden.

Practice your answers

Technical practice

How would you turn a supply-planning problem into an AI solution?

What to coverThe posting calls for predictive and prescriptive analytics in supply chain operations. Suggested answer focus: define the decision, available data and success measure before choosing a prediction, simulation or optimization approach.

How would you prepare large structured and unstructured datasets for modeling?

What to coverThe role includes collecting, cleaning and preprocessing data such as text, images and audio. Suggested answer focus: explain checks for completeness and consistency, how you would handle unsuitable records, and how preprocessing would differ by data type.

How would you tune a slow Spark SQL pipeline in Azure Databricks?

What to coverThe posting specifically names joins, partitions and file sizing as performance-tuning concerns. Suggested answer focus: describe how you would identify the bottleneck, compare changes using execution evidence and avoid sacrificing correctness for speed.

How would you orchestrate a Databricks workflow with Azure Data Factory?

What to coverThe qualifications name ADF pipelines, triggers, parameterization and integration runtimes, alongside ADLS Gen2 and Key Vault integration. Suggested answer focus: explain the responsibilities of each component and how you would handle failures and sensitive configuration.

How would you detect and repair a data-quality issue in a Delta Lake pipeline?

What to coverThe posting calls for data quality at scale, monitoring, alerting and debug/backfill playbooks. Suggested answer focus: discuss checks that expose the issue, how to trace affected data and how to backfill safely without creating inconsistent results.

How would you choose and evaluate a model for demand or item-level analytics?

What to coverThe role involves selecting among approaches such as regression, classification, NLP and computer vision. Suggested answer focus: tie the approach and evaluation measure to the business outcome, and explain how you would check that results hold beyond the training data.

How would you assess whether a prediction improves a planning or optimization decision?

What to coverThe posting covers supply-chain simulations and optimization as well as predictive analytics. Suggested answer focus: distinguish model accuracy from decision value, and describe how you would compare proposed actions under relevant constraints and scenarios.

How would you deploy and maintain a production AI model?

What to coverThe responsibilities include deployment through APIs or cloud platforms, performance monitoring and retraining when needed. Suggested answer focus: cover release checks, monitoring for degraded results and criteria for investigating or retraining rather than automatically replacing a model.

How would you make a Databricks and ADF solution reproducible across environments?

What to coverThe qualifications mention Git, reviews, tests, CI/CD and infrastructure as code using Terraform or Bicep. Suggested answer focus: explain how you would test code and pipeline changes, review releases and manage infrastructure changes consistently.

How would you evaluate an agentic AI capability for an operational workflow?

What to coverThe role explicitly includes implementing agentic AI and evaluating emerging AI architectures. Suggested answer focus: identify a bounded use case, compare it with a simpler approach, and propose ways to evaluate output quality, operational reliability and cost before deployment.

Behavioural practice

Tell me about a time you translated an ambiguous business need into a technical plan.

What to coverThe role works across data management, data science, decision science and technology on planning and optimization needs. Suggested answer focus: use a real situation, your actions to clarify the decision and align stakeholders, and the resulting outcome.

Describe a time you helped move an analytics prototype into production.

What to coverThe posting spans prototyping, deployment and ongoing operation, with an advisory role alongside architects and data scientists. Suggested answer focus: describe the situation, what you did to address production requirements with others, and the result.

Tell me about a production data or model issue you helped resolve.

What to coverMonitoring, alerting, debugging and backfills are stated parts of the work. Suggested answer focus: explain the situation, your role in diagnosis and communication, the corrective action and the outcome without assuming every issue required a backfill.

Describe a time a code review or stakeholder concern changed your implementation.

What to coverThe role calls for code reviews and collaboration with product, governance, security and technical stakeholders. Suggested answer focus: give a real situation, explain how you evaluated and acted on the feedback, and state the result.

Tell me about a time you balanced performance or cost against delivery needs.

What to coverThe qualifications name observability, cloud cost control and cost/performance trade-offs. Suggested answer focus: describe the constraints, the evidence behind your decision, the action you took and its measurable or observable result.

Sharpen your knowledge

Topics to prepare

Supply-chain problem framing
Source-backed: HAVI names demand and supply planning, replenishment, pricing, promotions and optimization. Revision exercise: take a hypothetical replenishment decision and distinguish the prediction needed from the decision to optimize, its inputs and how success would be measured.
Model and data choices
Source-backed: the role selects among classification, regression, NLP and computer vision, and handles structured and unstructured data. Revision exercise: compare two approaches for a hypothetical item-level analytics problem, explaining preprocessing, evaluation and when a simpler model would suffice.
Databricks and Delta Lake architecture
Source-backed: the qualifications emphasize Azure Databricks, PySpark or Scala, Spark SQL, Delta Lake and large-dataset tuning. Revision exercise: diagram a Delta Lake pipeline and explain how joins, partitions and file sizing affect its performance.
Azure orchestration and integration
Source-backed: the posting specifies Azure Data Factory pipelines, triggers, parameterization and integration runtimes, integrated with ADLS Gen2 and Key Vault. Revision exercise: trace how a parameterized pipeline would read data, obtain a secret and handle a failed run.
Production reliability and cost
Source-backed: HAVI calls for data-quality checks, monitoring, alerting, debug/backfill processes, observability and cloud cost/performance management. Revision exercise: outline failure detection, recovery and cost trade-offs for a recurring data-and-model workflow.
Emerging and agentic AI
Source-backed: the role includes evaluating transformers and GANs and implementing agentic AI capabilities. Suggested preparation, not a specified project: compare where an agentic workflow would add value against a conventional workflow, including how its outputs would be checked.

Skills to revise

Build a Spark-to-Delta pipeline
Suggested exercise: use sample structured data to clean and transform records with PySpark or Scala, query them with Spark SQL and write a Delta table. This practices the explicitly requested Azure Databricks stack.
Tune a large-data query
Suggested exercise: inspect a slow Spark SQL query, revise its joins, partitioning or file layout, and record the before-and-after plan and runtime. The posting explicitly calls for SQL performance tuning across large datasets.
Configure an orchestrated data flow
Suggested exercise: draft or build a parameterized Azure Data Factory pipeline with a trigger, an ADLS Gen2 input and a Key Vault-backed secret; explain the integration-runtime choice. These components are named in the qualifications.
Test data quality and recovery
Suggested exercise: add data-quality checks and an alert to a sample pipeline, then document how to debug and backfill one failed batch without duplicating records. HAVI explicitly expects quality at scale, monitoring, alerting and debug/backfill processes.
Operationalize a model
Suggested exercise: train a small regression or classification model, expose a prediction endpoint, define a performance check and describe a retraining trigger. This reflects the stated build, deploy, monitor and retrain responsibilities; the example problem is a preparation suggestion.
Practice release and infrastructure controls
Suggested exercise: put a sample pipeline under Git review with unit and integration tests, sketch a CI/CD release, and outline Terraform or Bicep resources and Azure Monitor checks. The posting names these practices, infrastructure as code, and cost/performance controls.

Your preparation checklist

Map the role to evidence
Create a one-page evidence map for the posting’s requirements, especially 2+ years of hands-on Azure Databricks, PySpark or Scala, Spark SQL, Delta Lake, Azure Data Factory, SQL tuning, and production support. Note which examples you can substantiate and where you would discuss an approach instead.
Prepare a supply-chain project walkthrough
Choose a relevant project, if available, and explain the business problem, stakeholders, data, model or optimization approach, deployment, and measured outcome. HAVI names demand and supply planning, replenishment, pricing, promotions, and item/menu analytics as application areas.
Work through a data-pipeline design
Sketch an approach for ingesting and preparing large supply-chain datasets with Azure Data Factory, ADLS Gen2, Key Vault, Databricks, and Delta Lake. Be ready to explain pipeline triggers, parameterization, access to secrets, and SQL performance choices such as joins, partitions, and file sizing.
Rehearse a model-to-production exercise
For a hypothetical demand-planning problem, outline how you would select and evaluate a model, expose it through an API or cloud deployment, monitor performance, and decide when to retrain. Explain where an agentic AI capability might help, without assuming HAVI uses a particular architecture.
Show how you would support a failing data product
Prepare a concise incident walkthrough covering data-quality checks, alerting, diagnosis, safe backfills, and validation. The posting specifically calls for large-scale data quality, monitoring, alerting, and debug/backfill playbooks.
Gather delivery and cost-control examples
If available, identify examples of Git-based reviews, unit or integration tests, CI/CD, and Terraform or Bicep. Prepare to explain observability and a concrete trade-off among cluster sizing, autoscaling, Photon, performance, and cloud cost.
Write questions for the interviewers
Ask which supply-chain use cases are priorities, how analytics and technology teams divide ownership from prototype through operations, and how model and data-product success is measured. You can also clarify expectations for the Chicago-based hybrid schedule.

Availability last checked: Nov 4, 2026. Status can change after a check; confirm on the employer’s website. Report an issue · Our checking policy

KEEP EXPLORING

Similar AI jobs

Related skills and specializations in United States. Matched to this role, not your profile.

Explore more jobs

Finding similar opportunities…