Lead AI/ML Engineer

What matters most

• Set direction and lead delivery

  The role owns the AI/ML and Generative AI roadmap, architecture guidance, code reviews and mentoring. It also calls for translating business problems with technical and non-technical stakeholders into scalable solutions.

• Build and evaluate enterprise GenAI systems

  The posting emphasizes AWS-based Generative AI, LLM fine-tuning, evaluation and deployment, prompt engineering, agentic workflows, and optimized RAG pipelines.

• Own production ML lifecycles

  Responsibility extends from experiments to deployment and monitoring, including automated preprocessing, training and validation, plus SageMaker model versioning, drift detection and retraining strategies.

• Deliver across ML and computer vision

  Beyond LLMs, the role covers supervised, unsupervised and deep learning; forecasting and predictive modeling; and deployed vision models for classification, detection, segmentation and image captioning.

• Connect models to cloud applications

  The requirements call for AWS services including SageMaker and Bedrock, REST API development, SQL and NoSQL databases, and caching—not just model development.

Likely challenges to think through

• Choosing the right GenAI approach

  A likely difficulty is deciding when prompting, RAG, agentic workflows or fine-tuning warrants the added complexity while maintaining useful evaluation. Prepare to explain decision criteria and how you would compare approaches; the posting does not specify a current project or preferred choice.

• Moving experiments into maintainable production

  Rapid prototyping can conflict with the effort needed for automated pipelines, versioning, monitoring and retraining. Prepare an example or design walkthrough that makes those handoffs and trade-offs explicit, without assuming ARMAKUNI INDIA has a particular deployment issue.

• Balancing technical depth with leadership breadth

  The remit spans LLMs, conventional ML, computer vision, cloud integration and team guidance, making prioritization a likely challenge. Prepare to explain how you would scope work, delegate reviews and align architecture decisions with stakeholder needs; these are preparation hypotheses, not confirmed team problems.

Topics to prepare

• Technical leadership and solution architecture

  The role owns the AI/ML roadmap, architecture decisions, code reviews and mentoring. Rehearse how you would turn a business problem into a scoped solution, explain trade-offs to technical and non-technical stakeholders, and review an implementation plan.

• LLM adaptation and evaluation

  The posting calls for LLM fine-tuning, evaluation, deployment and prompt engineering. Revise when prompting versus fine-tuning is appropriate, then sketch an evaluation plan with baseline results, test cases and failure analysis.

• Enterprise RAG and agentic design

  RAG pipelines, vector databases, indexing strategies and agentic workflows are explicit responsibilities. Compare two indexing approaches and trace how retrieval quality, prompt design and workflow steps affect an answer.

• Production ML lifecycle

  The role covers experimentation through monitoring, including SageMaker model versioning, drift detection and retraining. Draw a lifecycle diagram and identify the evidence you would check before deploying or retraining a model.

• Predictive ML and forecasting

  The posting specifies supervised, unsupervised and deep learning, plus time series forecasting. Revise suitable validation and evaluation choices for each; explain how a forecasting split avoids using future observations.

• Computer vision task design

  Classification, object detection, segmentation and image captioning are named. For each task, identify the required output and an appropriate way to evaluate it, then explain what would change at deployment.

Skills to revise

• Python model comparison

  As practice for the stated Python and ML-library requirements, build a small reproducible comparison using Scikit-learn and one of XGBoost, LightGBM or CatBoost; document preprocessing, validation and results.

• RAG workflow implementation

  As a preparation exercise, prototype retrieval and answer generation using a named framework such as LangChain or LlamaIndex and a named index such as FAISS or Chroma. Inspect retrieved passages and test how indexing changes results.

• AWS lifecycle walkthrough

  The posting names SageMaker, Bedrock, S3 and Lambda. Practice explaining or diagramming a prototype-to-deployment workflow that includes model versions, monitoring and a retraining decision; distinguish demonstrated experience from a proposed design.

• Vision model experiment

  Using PyTorch or TensorFlow, practice outlining a small classification or detection experiment, including data preparation, validation and inference. This targets the posting’s hands-on deep learning and computer vision work.

• API and data-serving design

  Sketch a REST inference endpoint using the posting’s FastAPI or Flask options, with Docker packaging. Explain where PostgreSQL or DynamoDB would fit and when Redis or Memcached caching would help; this is a design exercise, not a stated project.

• Architecture review communication

  Practice a short code-review explanation and a non-technical account of an AI solution’s trade-offs. Both exercises relate directly to the role’s mentoring, architecture guidance and stakeholder discussions.

• Build a dated evidence sheet

  List actual AI/ML and Generative AI/LLM work with dates, your responsibilities, Python tools and measurable outcomes. Use it to discuss the posting’s 6+ years of AI/ML and 3+ years of GenAI/LLM experience without overstating experience.

• Rehearse a technical leadership walkthrough

  Prepare one real example covering the business problem, technical roadmap, architecture decisions, code reviews or mentoring, stakeholder discussions and production outcome. Distinguish your decisions from the team’s work.

• Practice an enterprise RAG design exercise

  Sketch a hypothetical AWS-based solution covering document ingestion, indexing and vector-store choice, retrieval, prompting or agent workflow, LLM evaluation and monitoring. Explain when you would consider fine-tuning instead; label the design as practice, not a past deployment.

• Prepare an ML lifecycle case study

  Walk through a real supervised, unsupervised or forecasting project from preprocessing and validation to deployment and monitoring. Be ready to explain model versioning, drift detection and retraining, including SageMaker where you have used it.

• Prepare a computer vision example

  If you have relevant work, outline a classification, detection, segmentation or image-captioning project with its data, model choice, evaluation and deployment constraints. Otherwise, prepare a clearly hypothetical design discussion.

• Diagram a production integration

  Practice explaining how an AI service could expose a REST API, run in a Docker-based deployment and connect to appropriate AWS storage, compute, database or caching services named in the posting. State trade-offs rather than claiming ARMAKUNI INDIA uses that architecture.

• Draft questions for the employer

  Prepare questions about which AI/ML and GenAI problems take priority, how the team measures production success, what infrastructure is already in place and the lead’s scope for architecture, mentoring and roadmap decisions.