Senior Manager, AI/ML Engineering

What matters most

• Own the predictive-design roadmap

  The role owns ML systems that deliver genomic, phenotypic and multi-omic predictions to scientists, from problem framing through evaluation and delivery.

• Make predictions trustworthy

  Define model evaluation and validation against experimental and field outcomes, working with computational and crop biologists to interpret heterogeneous biological data.

• Ship operable ML systems

  Set technical standards and remain accountable for delivery, quality and operability, including deployment, maintenance and monitoring of production models and pipelines.

• Apply modern AI selectively

  Drive useful agentic, generative AI, RAG and foundation-model solutions while deciding when to build versus adopt models or third-party platforms.

• Lead across technical and scientific teams

  Manage and develop ML engineers, oversee focused contractor work, contribute hands-on where it matters most and represent the AI Team in Data & AI strategy discussions.

Likely challenges to think through

• Protect exploration without delaying delivery

  A likely trade-off, not a reported problem: production-ready workflows need dependable delivery while new approaches to crop-trait modeling need room to be tested. Prepare a way to prioritize experiments, set decision points and promote promising work into production.

• Connect model scores to scientific outcomes

  A likely difficulty is that performance on heterogeneous biological data may not establish value against experimental or field outcomes. Prepare to explain how you would agree on validation criteria with scientists and communicate uncertainty before predictions inform decisions.

• Choose where modern AI earns its complexity

  Build-versus-adopt decisions for foundation models, agents and platforms may pit speed against control, fit and operability. Prepare a comparison framework tied to scientific or business outcomes, including reasons not to use an agentic approach.

• Stay hands-on while scaling the team

  Technical contribution can compete with mentoring, contractor oversight and cross-functional leadership. Prepare an example of how you would reserve hands-on work for high-impact problems while delegating delivery and maintaining clear standards.

Topics to prepare

• Model evaluation against biological outcomes

  The role defines evaluation and validation against experimental and field outcomes. Revise how to select baselines, avoid misleading validation splits, and interpret disagreement between model predictions and observed results; sketch an evaluation plan for a hypothetical crop-trait predictor.

• Biological data and ML problem framing

  The roadmap serves genomic, phenotypic, and multi-omic predictions to scientists. Practice turning a hypothetical crop scientist’s trait question into a prediction target, available inputs, and success criteria. Familiarity with genomics, transcriptomics, and proteomics is preferred, not required.

• Production ML architecture and operability

  The posting requires production models and pipelines at scale, with ownership of delivery, quality, and operability. Diagram a model’s path from training to deployment and monitoring, then identify likely failure points and how the team would detect them.

• Foundation models, RAG, and agentic-system judgment

  Practical knowledge of these systems and decisions about building versus adopting them are required. Compare a hypothetical scientific information task with a prediction task; explain where RAG, an agent, a foundation model, or a simpler approach would fit—and where it would not.

• Technical roadmap and team leadership

  The manager sets standards, mentors ML engineers, and balances production delivery with modeling exploration. Draft a short roadmap for two hypothetical modeling opportunities, stating priorities, evaluation gates, ownership, and what would trigger a change in direction.

Skills to revise

• Train and deploy a neural-network example

  Using Python and PyTorch or TensorFlow, rehearse training a small model on synthetic data, recording evaluation results, and preparing it for deployment. This directly reflects the required hands-on neural-network and scientific-stack experience; the synthetic task is only a practice exercise.

• Trace an ML pipeline through the named stack

  Outline or build a small practice pipeline covering a Docker container, AWS or SageMaker deployment, MLflow experiment tracking, and Airflow scheduling, using equivalents where appropriate. Show how a failed run or degraded model would be detected and addressed; deployment, maintenance, and monitoring at scale are explicit requirements.

• Write a scientist-facing validation brief

  For a hypothetical trait-prediction model, produce a one-page brief defining the biological question, evaluation results, limitations, and what experimental or field evidence would be needed before acting. This practices the required translation between scientific partners and ML framing.

• Prototype and assess a RAG or agentic workflow

  On a small, nonproprietary document set, sketch or implement retrieval and an agentic step, then test incorrect retrieval, unsupported answers, and failure handling. The posting calls for applied RAG and agentic systems; this prototype is a preparation suggestion, not a stated assignment.

• Rehearse a cross-functional technical decision

  Prepare a five-minute explanation to scientists, engineers, and leaders of whether to build or adopt a hypothetical modeling capability, including quality criteria, operational trade-offs, and delivery priorities. This matches the role’s roadmap ownership, leadership discussions, and cross-functional communication.

• Map your evidence to the required qualifications

  Prepare a one-page reference to your education or equivalent experience, years in ML or AI roles, production models and pipelines, Python and neural-network work, and team leadership. Mark which examples you can substantiate; the posting requires these areas.

• Rehearse a production ML project walkthrough

  Prepare one end-to-end example covering problem framing, your hands-on contribution, model training, deployment, monitoring, failures, and measured outcomes. Be ready to explain the tools you used and how they relate to PyTorch or TensorFlow, AWS, SageMaker, Docker, MLflow, and Airflow or equivalents named in the posting.

• Sketch a predictive-design system

  As interview practice, draw a hypothetical system that turns genomic, phenotypic, and multi-omic data into predictions scientists can use. State assumptions, data and pipeline boundaries, serving choices, and how computational biologists, data engineers, software engineers, and crop scientists would contribute; do not present the sketch as Inari’s architecture.

• Draft a model-validation plan

  Prepare a proposed evaluation approach that connects model metrics to experimental and field outcomes, addresses heterogeneous biological data, and explains when scientists should or should not trust a prediction. Label the approach as your proposal, not an existing Inari process.

• Write a build-versus-adopt decision example

  Prepare a brief comparison for a possible foundation-model, RAG, or agentic use case: the scientific or business need, simpler alternatives, evaluation criteria, operational risks, and reasons to build, adopt, or decline it. The role explicitly calls for this judgment.

• Prepare a leadership and roadmap story

  Select an example showing how you set technical standards, prioritized production delivery against exploration, mentored engineers, and influenced partners or leaders. If relevant, explain how you scoped and oversaw a focused contractor engagement while remaining hands-on.

• Bring questions that test the role’s priorities

  Write questions about the first modeling priorities, how predictions are checked against experimental and field results, where agentic systems could help scientists, and how the AI Team works with biological and engineering partners. These are questions to ask, not assumptions about Inari’s current processes.