New opportunity

AI Infrastructure Architect

Accenture · Gurugram, India

About this role

Accenture seeks an AI Infrastructure Architect in Gurugram to design and build AWS-based infrastructure for large-scale AI and machine learning systems. The role covers distributed training, accelerated compute, model serving and production operations, with attention to performance, power consumption, scalability, security and cost.

Responsibilities

Architect and tune GPU clusters, container platforms, secure networking and high-throughput storage using AWS services including EC2, EKS, SageMaker, S3, FSx/EFS, VPC, IAM and CloudWatch. Evaluate technology, vendors and architecture alternatives across compute, storage, orchestration and operations.
Review environments for risks and bottlenecks; document architecture decisions, standards and remediation plans. Develop capacity plans, scaling strategies, cost forecasts and infrastructure roadmap inputs.
Design automation and CI/CD approaches for AI systems, models, data pipelines and platform components. Establish monitoring, alerting, SLAs and SLOs across infrastructure and MLOps, and support compliance with client and regulatory requirements.
Work with clients, architects and engineers to translate business requirements into infrastructure designs; lead workstreams, review designs and code, and mentor engineers.

Requirements

The posting specifies a minimum of 12 years of experience, while its job header lists 10–12 years. It also requires a bachelor's degree in computer science, computer engineering, information technology or a related engineering field, and states 15 years of full-time education. Required experience includes at least four years building or operating AI/ML infrastructure or related large-scale platforms, at least four years in AI/ML infrastructure engineering or related roles, and at least four years of programming or scripting proficiency. Candidates need AWS AI infrastructure expertise, distributed training and model-serving experience, familiarity with workflow tools such as Airflow or Kubeflow, infrastructure automation, observability, and the ability to lead engineering work and communicate with stakeholders.

Preferred

AWS professional or AI/ML certifications; experience in regulated or other industry environments; exposure to LLM infrastructure, vector databases, retrieval pipelines, GPU scheduling and model optimization; and knowledge of enterprise architecture governance, FinOps and vendor collaboration.

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

Skills for this role

AI infrastructure architectureAWSEC2EKSSageMakerS3Amazon FSxAmazon EFSVPCIAMCloudWatchGPU computingDistributed trainingModel servingKubernetesDockerTerraformCloudFormationCI/CDPythonJavaC++BashPowerShellApache AirflowKubeflowMLOpsInfraOpsObservabilityIncident response

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

Architect AWS infrastructure for training and serving
The posting calls for end-to-end designs for GPU clusters, distributed training, container platforms and model serving, including accelerator, networking and high-throughput storage choices.
Make defensible architecture decisions
The role evaluates alternatives across performance, power, scalability, security, cost and operational complexity. It also documents decisions, assumptions and remediation plans after architecture reviews.
Build repeatable production operations
Responsibilities include automation and CI/CD for AI systems, models and data pipelines, plus monitoring, alerting, SLAs, SLOs and performance and cost tracking.
Align designs and lead delivery
The architect translates client requirements into designs that meet standards and compliance needs, contributes capacity plans and cost forecasts, leads workstreams, reviews designs and code, and mentors engineers.

Likely challenges to think through

Balancing accelerator performance against cost and power
Preparation hypothesis, not a confirmed employer problem: faster GPU configurations may increase spend and power consumption. Prepare a decision example that compares workload needs, utilization, throughput and cost, and explains what evidence would justify the choice.
Meeting different training and serving demands
Preparation hypothesis: distributed training can strain networking and storage, while model serving introduces latency and reliability targets. Prepare to explain how you would size each workload, identify bottlenecks and revise capacity plans as demand changes.
Keeping releases repeatable without weakening controls
Preparation hypothesis: rapid model and pipeline releases can conflict with client security, compliance and operational requirements. Prepare an approach to automated deployment, access controls, observability and rollback that makes those trade-offs explicit.

Practice your answers

Technical practice

How would you design and tune an AWS GPU cluster for distributed training?

What to coverSuggested answer focus: Explain how workload characteristics guide accelerator selection, interconnect and network design, and training scalability. Discuss how you would measure utilization, performance, power consumption and cost before changing the design.

How would you choose storage for a high-throughput AI training workload?

What to coverSuggested answer focus: Compare the roles of S3 and FSx or EFS using data access patterns, throughput, latency and operational cost. Explain how you would diagnose a storage bottleneck rather than assuming more GPU capacity is needed.

How would you design a reliable model-serving foundation?

What to coverSuggested answer focus: Cover deployment, scaling, latency, availability and monitoring for production inference. Explain the trade-offs between performance and cost, and how service objectives would influence the design.

When would you choose EC2, EKS or SageMaker for an AI workload?

What to coverSuggested answer focus: Compare control over infrastructure, orchestration needs, operational complexity and cost across the options. State what workload and client constraints you would establish before recommending an architecture.

How would you secure AI infrastructure within an enterprise AWS environment?

What to coverSuggested answer focus: Address VPC network boundaries, IAM access, data protection and integration with client standards. Explain how you would turn compliance or regulatory requirements into reviewable design decisions without assuming a particular regulation applies.

How would you make AI infrastructure and releases repeatable?

What to coverSuggested answer focus: Describe an infrastructure-as-code and CI/CD approach spanning platform components, models and data pipelines, using Terraform or CloudFormation where appropriate. Discuss validation, controlled rollout and recovery when a change fails.

How would you integrate workflow tooling with AI infrastructure?

What to coverSuggested answer focus: Discuss how Airflow, Kubeflow or platform-native orchestration could coordinate data and model workflows, including dependencies and failure handling. The posting also names AI Agents & Workflow Integration as a must-have skill; as a preparation suggestion, consider how you would clarify an agent workflow's infrastructure, access and observability needs before designing for it.

How would you establish observability and respond to a production AI incident?

What to coverSuggested answer focus: Connect CloudWatch and other infrastructure or MLOps signals to SLIs, SLOs, alerts and runbooks. Walk through how you would isolate a training or serving bottleneck, restore service and use the findings for improvement.

How would you assess an existing AI environment and document an architecture choice?

What to coverSuggested answer focus: Explain how you would identify risks, bottlenecks and gaps, then compare remediation options across performance, security, cost and operational complexity. Describe what assumptions, dependencies and trade-offs belong in an architecture decision record.

How would you plan capacity and forecast costs for growing AI workloads?

What to coverSuggested answer focus: Relate demand forecasts to accelerator capacity, storage, scaling strategies and service objectives. Explain how you would test assumptions, track actual performance and spending, and revise roadmap recommendations.

Behavioural practice

Tell me about a time you translated unclear stakeholder needs into an infrastructure design.

What to coverSuggested answer focus: Use a real situation, the questions you asked clients or architects, and the decisions you documented. Explain the resulting alignment or design outcome without claiming a specific result you did not achieve.

Describe a time you led an engineering workstream with competing priorities.

What to coverSuggested answer focus: Give the situation, how you set direction and managed trade-offs, and what resulted. Make clear which decisions you owned and how you kept others informed.

Tell me about a time a design or code review improved an engineer's work.

What to coverSuggested answer focus: Describe the context, the feedback or mentoring action you took, and the result. Highlight how you maintained standards while enabling the engineer to make progress.

Describe a production issue you helped troubleshoot across teams.

What to coverSuggested answer focus: State the operational impact, your diagnostic and communication actions, and the outcome. Include what you changed afterward in monitoring, runbooks or design, if applicable.

Tell me about a time stakeholders disagreed on an architecture trade-off.

What to coverSuggested answer focus: Explain the competing goals, the evidence you used to evaluate options, and how a decision was reached. Describe the result and how you recorded assumptions or remaining risks.

Sharpen your knowledge

Topics to prepare

Distributed training and accelerated compute
Compare two GPU-cluster designs on paper, explaining accelerator selection, interconnects, storage throughput and the trade-offs among performance, power, scale and cost. The role explicitly covers large-scale distributed training and GPU infrastructure.
AWS AI infrastructure architecture
Map how EC2, EKS, SageMaker, S3, FSx/EFS, VPC, IAM and CloudWatch could support a training-to-serving architecture. Explain why each service belongs in the design and where alternatives might be preferable; these services appear in the posting.
Model serving and workflow integration
Trace a model from data pipeline through deployment to production serving, identifying orchestration, scaling and failure points. The posting requires model-serving experience and familiarity with workflow tools such as Airflow or Kubeflow; its header also lists AI Agents & Workflow Integration as a must-have skill.
Production reliability and observability
Define sample SLIs, SLOs, alerts and incident priorities for training and inference workloads. Relate them to the posting’s requirements for InfraOps and MLOps monitoring, SLAs, incident response and performance/cost tracking.
Security, compliance and architecture trade-offs
Review a hypothetical client design for access, network isolation, compliance risks, capacity and cost; write down assumptions and remediation options. The role calls for secure AWS designs, architecture reviews, forecasts and documented decisions.

Skills to revise

Produce an architecture decision record
Draft a one-page decision comparing two AWS compute-and-storage approaches, with assumptions, bottlenecks, cost implications and a recommendation. The role calls for defensible architecture decisions, standards and remediation plans.
Write infrastructure automation
Sketch a small Terraform or CloudFormation configuration for network, compute and access components, then review it for repeatability and security. Both tools are named in the required skills.
Design a deployment workflow
Draw a CI/CD sequence for a model and its platform components, including validation, rollout and rollback; show where Airflow or Kubeflow could orchestrate a data or training workflow. The posting calls for production deployment automation and workflow-tool experience.
Script an operational check
Write a short Python or Bash example that checks workload status or summarizes performance metrics, and explain how it would surface a failure. Programming or scripting proficiency and operational troubleshooting are explicit requirements.
Practice an incident review
Using a hypothetical model-serving latency spike, outline CloudWatch signals, triage steps, escalation and a runbook update. This exercises the posting’s observability, incident-response and production-reliability expectations.
Present a capacity recommendation
Create a brief capacity and cost estimate for a growing training or serving workload, state its assumptions, and explain the recommendation to a non-specialist stakeholder. Capacity planning, cost forecasts and client communication are part of the role.

Your preparation checklist

Build an AWS architecture walkthrough
Prepare a diagram and decision record for a distributed-training and model-serving environment. The posting names EC2, EKS, SageMaker, S3, FSx/EFS, VPC, IAM and CloudWatch; explain service choices, assumptions and alternatives.
Practice a GPU cluster design exercise
Work through accelerator selection, networking, high-throughput storage and scaling for a hypothetical workload. Show how you would test performance, power consumption, reliability and cost trade-offs without inventing target figures.
Prepare a production operations example
Outline an example covering infrastructure automation, CI/CD, workflow orchestration, monitoring, SLOs, alerting and incident response. Distinguish what was implemented from what you would propose.
Draft an architecture review and remediation plan
Create a sample assessment of bottlenecks, security and compliance risks, with prioritized actions, capacity assumptions and a cost-forecast approach. The posting explicitly calls for reviews, documented decisions and roadmap inputs.
Select evidence of technical and workstream leadership
Identify projects that can demonstrate programming or scripting, AI/ML infrastructure engineering, technology evaluation, design or code review, mentoring and stakeholder communication. Prepare precise accounts of your role and outcomes.
Prepare questions to clarify the role
Ask how the listed must-have skill, AI Agents & Workflow Integration, relates to the AWS infrastructure responsibilities; which workloads and client standards are in scope; and how the header’s 10–12 years relates to the stated 12-year minimum.

Availability last checked: 5 Nov 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 India. Matched to this role, not your profile.

Explore more jobs

Finding similar opportunities…