New opportunity

AI Infrastructure Architect

Accenture · Gurugram, India

About this role

Architect and build AWS-based compute infrastructure for large-scale AI and machine learning systems, optimizing performance, power consumption, cost and scalability. The role covers distributed training, GPU clusters, container platforms, model serving and integration with enterprise systems. The posting identifies AI Agents & Workflow Integration as a must-have skill.

Responsibilities

Design and tune accelerated compute, networking, high-throughput storage and model-serving environments using AWS services including EC2, EKS, SageMaker, S3, FSx/EFS, VPC, IAM and CloudWatch.
Assess architecture alternatives and existing environments; identify risks and bottlenecks, document decisions and recommend improvements across security, compliance, reliability, cost and operational complexity.
Develop infrastructure roadmaps, capacity and scaling plans, cost forecasts, deployment automation and CI/CD approaches for AI systems, models and data pipelines.
Establish monitoring, alerting, SLAs and SLOs across InfraOps and MLOps. Collaborate with clients and engineers, lead workstreams, review designs and code, and mentor team members.

Requirements

The job description states a minimum of 12 years of experience; the listing header separately says 10–12 years. A bachelor's degree in Computer Science, Computer Engineering, Information Technology or a related engineering field and 15 years of full-time education are specified.
At least 4 years building and operating AI/ML infrastructure, cloud or data platforms, model deployment pipelines or comparable large-scale solutions; at least 4 years in AI/ML infrastructure engineering or related hyperscaler or enterprise-platform roles; and at least 4 years of programming or scripting proficiency. These are overlapping, role-specific requirements, not additional years to add to the overall minimum.
Hands-on AWS AI infrastructure experience; GPU computing, distributed training, model serving, secure networking, containers, Terraform or CloudFormation, CI/CD, Docker, Kubernetes, observability and incident response. Experience with workflow tools such as Airflow or Kubeflow, technical leadership, technology evaluation and stakeholder communication is required.

Preferred

AWS certifications; experience in regulated or cost-sensitive industries; exposure to LLM infrastructure, vector databases, retrieval pipelines, GPU scheduling and model optimization; and familiarity with 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 Agents & Workflow IntegrationAmazon Web Services (AWS)GPU computingDistributed trainingModel servingAmazon EC2Amazon EKSAmazon SageMakerAmazon S3Amazon FSxAmazon EFSAmazon VPCAWS IAMAmazon CloudWatchPythonJavaC++BashPowerShellApache AirflowKubeflowTerraformAWS CloudFormationDockerKubernetesCI/CDMLOpsInfraOpsObservabilityIncident 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

AWS AI infrastructure architecture
Source-backed: Own the design of large-scale AI/ML infrastructure spanning GPU compute, distributed training, container platforms and model serving. Prepare to explain AWS design choices across EC2, EKS, SageMaker, networking and storage, including why alternatives were rejected.
Defensible performance and cost decisions
Source-backed: Tune accelerator, network and high-throughput storage designs while weighing performance, power, scalability, cost, security and operational complexity. Prepare a concrete example of an architecture decision with assumptions, measured bottlenecks and trade-offs.
Production operations and repeatable delivery
Source-backed: Define capacity plans, cost forecasts, CI/CD and automation strategies, plus monitoring, alerting, SLAs and SLOs across InfraOps and MLOps. Prepare to trace a workload from deployment through observability, incident response and scaling.
Enterprise integration and technical leadership
Source-backed: Integrate AI/ML systems with client environments, align designs with security and compliance standards, lead workstreams and mentor engineers. AI Agents & Workflow Integration is the listed must-have skill, though the posting gives no agent-specific duties. Prepare to explain how workflow integration fits within a secure, operable architecture without assuming a particular agent framework.

Likely challenges to think through

GPU throughput versus infrastructure spend
Preparation hypothesis, not a confirmed employer problem: Faster training may call for more accelerators, network capacity and storage throughput, increasing cost and power use. Prepare a comparison of viable designs using workload assumptions, utilization, bottleneck measurements and scaling thresholds.
Reliability versus deployment complexity
Preparation hypothesis, not a confirmed employer problem: Production model serving and training pipelines need repeatable releases and useful alerts, but additional controls can complicate operations. Prepare to explain how CI/CD, SLOs, monitoring and runbooks would address failures without creating excessive operational overhead.
Client standards versus architecture flexibility
Preparation hypothesis, not a confirmed employer problem: A technically attractive design may conflict with enterprise security, compliance or existing-system constraints. Prepare an example of documenting options, risks and dependencies, then aligning stakeholders on a defensible decision.

Practice your answers

Technical practice

Suggested practice question: How would you architect an AWS GPU cluster for distributed training?

What to coverThe posting calls for GPU and accelerator selection, distributed training, networking, storage and scalable compute. Discuss how you would compare EC2, EKS and SageMaker options against workload performance, reliability, power and cost goals.

Suggested practice question: How would you find and resolve a data bottleneck in distributed training?

What to coverThe role calls for high-throughput storage and interconnect design using services including S3 and FSx/EFS. Explain what you would measure across data access, networking and accelerator utilization before changing the architecture.

Suggested practice question: How would you design a production model-serving foundation?

What to coverAddress serving capacity, scalability, latency, reliability and operational controls, all relevant to the advertised work. Explain how workload requirements would guide your choice of deployment approach rather than assuming one AWS service fits every model.

Suggested practice question: How would you integrate AI agents and workflows into an enterprise AI environment?

What to coverAI Agents & Workflow Integration is listed as a must-have skill, and the posting names Airflow, Kubeflow or comparable workflow tooling. As a preparation suggestion, describe how you would establish workflow interfaces, access controls and observability without assuming a particular agent framework is required.

Suggested practice question: How would you make AI platform and model releases repeatable?

What to coverThe role includes deployment automation and CI/CD for AI systems, models, data pipelines and platform components. Compare where Terraform or CloudFormation, containers and release checks would help, and how you would handle rollback and configuration changes.

Suggested practice question: How would you secure AWS AI infrastructure connected to enterprise systems?

What to coverThe posting names VPC and IAM and requires interoperability, security and compliance alignment. Discuss network boundaries, access design and how you would validate the architecture against client standards without assuming a specific regulatory regime.

Suggested practice question: How would you plan capacity and forecast the cost of growing AI workloads?

What to coverThe role calls for scaling plans, cost forecasts and optimization of performance, power consumption and cost. Explain which workload assumptions and utilization measures you would use, and how you would communicate trade-offs when demand changes.

Suggested practice question: How would you define observability and troubleshoot a production AI infrastructure incident?

What to coverThe posting calls for CloudWatch, InfraOps and MLOps monitoring, alerting, SLAs, SLOs and incident response. Describe signals that would distinguish infrastructure, pipeline and serving problems, then explain how findings would inform alerts and runbooks.

Suggested practice question: How would you assess an existing AI environment and document an architecture decision?

What to coverThe role requires identifying risks, bottlenecks and improvement opportunities and recording rationale, assumptions and dependencies. Walk through how you would compare alternatives across reliability, security, compliance, cost and operational complexity before recommending remediation.

Suggested practice question: When would you consider retrieval or vector-database infrastructure for an LLM workload?

What to coverLLM infrastructure, vector databases and retrieval pipelines are listed as good-to-have exposure, not core requirements. As an optional preparation suggestion, explain what workload evidence would justify these components and what serving, storage and operating costs they add.

Behavioural practice

Suggested practice question: Tell me about aligning stakeholders on a difficult infrastructure trade-off.

What to coverUse a real situation, your actions and the result to show how you translated business requirements into an architecture decision. Focus on communicating trade-offs such as performance, security, reliability and cost; do not assume all stakeholders initially agreed.

Suggested practice question: Describe how you led an engineering workstream while priorities competed.

What to coverUse situation, action and result to explain how you set technical direction, decided what to address first and coordinated delivery. The posting specifically calls for workstream leadership and managing priorities across initiatives.

Suggested practice question: Give an example of mentoring someone or improving a design through review.

What to coverDescribe the situation, the feedback or guidance you gave and the resulting change. Connect the example to the role’s design and code reviews, mentoring and promotion of engineering practices.

Suggested practice question: Tell me about responding to an infrastructure or production reliability issue.

What to coverUse situation, action and result to distinguish your troubleshooting contribution from the broader team’s work. Explain how you communicated impact, addressed the issue and used what you learned to improve monitoring or operations.

Suggested practice question: Describe a time you evaluated competing technologies or architecture patterns.

What to coverUse situation, action and result to show how you gathered evidence, weighed risks and documented a recommendation. The role explicitly involves technology evaluation, architecture decisions and communicating rationale to stakeholders.

Sharpen your knowledge

Topics to prepare

AI agents and workflow integration
Posting: AI Agents & Workflow Integration is the must-have skill; the posting also calls for pipeline and workflow management experience. Practice suggestion: map how an agent-driven workflow could connect to a model-serving endpoint and an enterprise system, marking security and failure-handling decisions as assumptions. The posting does not specify an agent framework.
Distributed training and GPU architecture
Posting: The role designs AWS GPU clusters and distributed training environments, including accelerator selection, interconnects and high-throughput storage. Practice suggestion: compare two training architectures and explain likely compute, network and storage bottlenecks.
AWS security and enterprise integration
Posting: Infrastructure decisions must address secure networking, IAM, interoperability, client standards and compliance. Practice suggestion: trace access and data flows through a proposed VPC-based AI architecture and identify controls and unresolved compliance assumptions.
Model serving and production releases
Posting: Model-serving foundations and repeatable CI/CD releases of models, pipelines and platform components are responsibilities. Practice suggestion: outline a serving architecture and explain how a release could be validated, monitored and rolled back.
Architecture trade-offs and operational targets
Posting: The architect weighs performance, power, cost, scalability, reliability and operational complexity, and defines capacity plans, forecasts, SLAs and SLOs. Practice suggestion: compare two capacity options against stated assumptions and identify the measurements needed to defend a recommendation.

Skills to revise

Design an AWS AI infrastructure blueprint
Posting: Hands-on design spans EC2, EKS, SageMaker, S3, FSx/EFS, VPC, IAM and CloudWatch. Practice suggestion: draw a training-to-serving architecture using only services justified by the workload; annotate networking, storage throughput and scaling choices.
Draft repeatable deployment automation
Posting: Terraform or CloudFormation, CI/CD and programming or scripting proficiency are required. Practice suggestion: sketch an infrastructure definition and release pipeline, then write a small Python or Bash validation script. Explain what each stage checks rather than assuming a particular client toolchain.
Model a workflow integration
Posting: Experience with tools such as Apache Airflow or Kubeflow is required alongside AI Agents & Workflow Integration. Practice suggestion: draft a workflow that starts a data or model task, handles a failure and records its outcome; explain where an agent interaction would fit if applicable.
Troubleshoot with operational signals
Posting: CloudWatch, InfraOps/MLOps observability, alerting and incident response are named. Practice suggestion: for a hypothetical rise in training time or serving latency, specify metrics, an SLO, an alert and a short investigation runbook.
Defend an architecture decision
Posting: The role documents architecture decisions, reviews risks and designs, and aligns clients and engineers. Practice suggestion: write a one-page decision record comparing two AWS designs, with assumptions, cost and reliability trade-offs, risks and a recommendation suitable for stakeholder review.

Your preparation checklist

Map the stated qualifications
Prepare a brief evidence sheet for the degree and 15 years of full-time education, the stated 12-year minimum, and the overlapping four-year experience requirements in infrastructure and programming. Note that the listing header separately says 10–12 years.
Rehearse an end-to-end project walkthrough
Prepare one relevant AI/ML infrastructure example, if available, showing your architecture decisions, implementation, scale, measured outcomes, and your own leadership contribution. The posting emphasizes AWS, production operation and workstream ownership.
Sketch an AWS AI infrastructure design
Draft a design exercise covering GPU compute, distributed training, networking, high-throughput storage, EKS or SageMaker, model serving, and VPC/IAM controls. Label assumptions and explain how the design addresses performance, power, cost and scalability.
Write a sample architecture decision record
Compare two plausible compute or serving approaches against security, compliance, reliability, cost and operational complexity. Include bottlenecks, dependencies, a capacity estimate and the evidence needed to validate your recommendation.
Prepare an automation and workflow example
Outline how infrastructure as code, containers, CI/CD and an Airflow, Kubeflow or comparable workflow would support repeatable AI releases. The posting names AI Agents & Workflow Integration as must-have; prepare a specific example if you have one, rather than assuming what implementation the employer uses.
Build an operations scenario
Prepare a sample dashboard and incident runbook with training or serving metrics, alerts, SLAs/SLOs, cost tracking, escalation and remediation. Explain how InfraOps and MLOps responsibilities would meet at a production incident.
Practice a stakeholder decision discussion
Rehearse how you would present trade-offs to a client, resolve competing priorities, document a decision, review an engineer’s design and mentor the team. These activities are stated responsibilities of the role.
Bring targeted questions for the employer
Ask which AI workloads and agent-workflow integrations are in scope, what client security and compliance constraints apply, how architecture ownership is divided, and how the 10–12-year listing header 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…