New opportunity

AI Operations Engineer

Cisco · Bangalore, India

About this role

Cisco’s Commerce Operations Engineering Center (COEC) is hiring an AI Operations Engineer in Bangalore, India, for a hybrid, full-time role. Working with senior engineers and AI, engineering, operations, and business teams, the engineer will build and support AI agents that automate Commerce Operations tasks, reduce manual effort, and improve speed, accuracy, and user experience.

Responsibilities

Develop agent workflows using LLM reasoning, tool and function calling, retrieval, and agent-to-agent communication. Build RAG pipelines over documents, structured data, and enterprise knowledge repositories.
Integrate agents with commerce platforms and enterprise data through secure REST and GraphQL APIs, webhooks, and event-driven services. Apply authentication, rate limiting, validation, and error handling; support A2A and MCP integrations.
Implement defined guardrails, human approval steps, LLM routing, and failover mechanisms. Use prompt and context engineering to improve outcome accuracy, safety, and consistency.
Contribute to development, testing, deployment, monitoring, and continuous improvement, including test datasets, tracing, audit logs, quality regression detection, CI/CD, rollback, and cost tracking. Troubleshoot issues, report efficiency, cycle-time, incident, and accuracy metrics, and document designs, integrations, tests, and procedures.

Minimum qualifications: At least 3 years of experience in software engineering, AI engineering, data engineering, or a related field; experience building LLM-powered, machine-learning, or automation applications; Python proficiency; and experience with an LLM or agent framework such as LangChain, LangGraph, AutoGen, CrewAI, OpenAI SDK, or Anthropic SDK. Working knowledge of RAG, prompt engineering, tool or function calling, and workflow orchestration is required, along with API or system integration experience, familiarity with Git, testing, debugging, code reviews, and CI/CD, and effective technical problem-solving and collaboration.

Preferred qualifications

Familiarity with A2A, MCP, OpenAPI-based tool integration, multi-agent systems, human-in-the-loop workflows, and guardrails; evaluation and observability tools such as LangSmith or Langfuse; vector databases and retrieval techniques; AI governance, access control, and PII handling; AWS, GCP, or Azure; MLOps or LLMOps; and commerce or supply-chain operations. A bachelor’s or master’s degree in a relevant field, or equivalent practical experience, is preferred. The position is listed as 40 hours per week.

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

Skills for this role

AI agentsPythonLLMsLangGraphLangChainAutoGenCrewAIOpenAI SDKAnthropic SDKA2AModel Context Protocol (MCP)RESTGraphQLWebhooksEvent-driven servicesRetrieval-augmented generation (RAG)Prompt engineeringContext engineeringTool callingFunction callingWorkflow orchestrationAPI integrationAuthenticationCI/CDGitTestingDebuggingCode reviewsLLM evaluationObservability of AI systems and applications (AI observability)

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

Automate Commerce Operations workflows
Develop and enhance AI agents for manual and repetitive tasks, using LLM reasoning, tool calling, retrieval, and workflow orchestration to improve speed, accuracy, and user experience.
Integrate agents securely with enterprise systems
Connect agents to commerce platforms and data sources through REST, GraphQL, webhooks, and event-driven services, with authentication, rate limiting, validation, and error handling.
Ground agent responses in enterprise knowledge
Build and maintain RAG pipelines over documents, structured data, and knowledge repositories; use prompt and context engineering to improve outcome accuracy and consistency.
Control automated actions
Implement defined guardrails, human approval steps, LLM routing, and failover mechanisms, while supporting agent-to-agent integrations such as A2A and MCP.
Evaluate and operate solutions after deployment
Contribute to test datasets, tracing, audit logs, regression detection, CI/CD, monitoring, rollback, and cost tracking. Troubleshoot issues and report operational measures such as cycle time, incidents, and accuracy.

Likely challenges to think through

Balancing automation with oversight
Possible challenge, not a confirmed Cisco problem: approvals and guardrails can slow workflows, while unchecked agent actions can reduce safety or accuracy. Prepare to explain where an agent should act, request approval, fail over, or stop, and how those choices would be tested.
Maintaining reliable grounding
Possible challenge, not a confirmed Cisco problem: documents and structured data may give an agent incomplete or conflicting context. Prepare a RAG evaluation approach that checks retrieval relevance and answer accuracy, then uses failures to refine retrieval or context.
Handling integration failures without losing control
Possible challenge, not a confirmed Cisco problem: enterprise APIs and event-driven services can fail, reject requests, or impose rate limits. Prepare to discuss validation, authentication, error handling, tracing, and recovery without duplicating automated actions.
Showing operational value while managing cost
Possible challenge, not a confirmed Cisco problem: improvements in cycle time or manual effort must be weighed against accuracy, incidents, and LLM cost. Prepare to describe baseline metrics, regression tests, production monitoring, and rollback criteria.

Practice your answers

Technical practice

How would you structure a Python agent workflow for a repetitive Commerce Operations task?

What to coverSuggested approach: Describe how an agent framework could coordinate reasoning, retrieval, tool calls, and explicit workflow steps. Explain where you would make behavior deterministic and how you would test the workflow before it acts on an enterprise system.

How would you build a RAG pipeline across documents, structured data, and enterprise knowledge repositories?

What to coverSuggested approach: Cover ingestion, access-aware retrieval, context assembly, and checks that an answer is grounded in the retrieved material. Discuss how you would investigate an inaccurate answer before changing the prompt or retrieval pipeline.

When would you use semantic search, hybrid search, or reranking in a RAG workflow?

What to coverSuggested approach: Compare their relevance, latency, and implementation trade-offs using examples of document and structured-data queries. These retrieval methods are preferred knowledge in the posting, not stated requirements for every workflow.

How would you make an agent's REST, GraphQL, webhook, or event-driven integration reliable and secure?

What to coverSuggested approach: Explain authentication, input and output validation, rate limiting, and error handling around tool calls. Discuss how you would prevent a failed or repeated integration call from silently producing an incorrect operational outcome.

What would you check when connecting agents or tools through A2A, MCP, or OpenAPI-based interfaces?

What to coverSuggested approach: Explain how you would verify interface definitions, permitted actions, data passed between components, and failure behavior. Distinguish the posting's work supporting A2A and MCP integrations from its preferred familiarity with OpenAPI-based tool integration.

How would you implement guardrails and human approval in an agent workflow?

What to coverSuggested approach: Identify actions that should pause for approval and describe validation, access controls, audit records, and PII handling around them. Explain how a defined guardrail should behave when an action is unsafe or its outcome is uncertain.

How would you improve inconsistent agent outcomes using prompt and context engineering?

What to coverSuggested approach: Discuss clearer instructions, relevant context, and constraints on tool use, then explain how you would compare results against test cases. Address accuracy and safety without assuming that prompt changes alone can fix retrieval or integration failures.

How would you evaluate an agent update and detect a quality regression?

What to coverSuggested approach: Describe representative test datasets, outcome checks, tracing, and audit logs that make failures diagnosable. Compare the update with a prior version on quality and operational measures rather than relying only on a few successful examples.

How would you deploy and safely reverse an update to an AI workflow?

What to coverSuggested approach: Connect Git, testing, code review, CI/CD, versioning, monitoring, and rollback to the solution lifecycle described in the posting. Explain what evidence would prompt a rollback and how you would track cost alongside quality.

How would you troubleshoot a slow or inaccurate production agent and report its impact?

What to coverSuggested approach: Trace the workflow to separate retrieval, model, tool, and integration issues, then verify a fix with tests and monitoring. Relate technical findings to the posting's measures, such as cycle time, incidents, operational accuracy, and efficiency gains.

Behavioural practice

Tell me about a time you translated an operational need into an automation solution.

What to coverSuggested approach: Use a real situation, your actions to clarify the task and its constraints, and the result. Explain how you checked that the solution addressed the users' actual workflow rather than just automating an isolated step.

Describe a time you coordinated with business and engineering teams on a system integration.

What to coverSuggested approach: Give the situation, what you did to resolve differing requirements or dependencies, and the result. Make your own contribution clear while recognizing the roles of the other teams.

Tell me about a difficult production issue you investigated.

What to coverSuggested approach: Set out the symptoms and impact, the evidence you used to narrow down the cause, the actions you took, and the result. Explain what you changed or documented to make a recurrence easier to detect or handle.

Describe a time code review or technical guidance changed your implementation.

What to coverSuggested approach: Explain the situation, the feedback, how you evaluated and applied it, and the outcome. The role works under senior engineers and technical leads, so focus on sound technical judgment and collaboration rather than claiming sole ownership.

Tell me about a time you measured and communicated the effect of an automation change.

What to coverSuggested approach: Describe the initial problem, how you selected and measured a meaningful outcome, what you communicated, and the result. If applicable, distinguish observed changes in speed, accuracy, incidents, or manual effort from effects you could not verify.

Sharpen your knowledge

Topics to prepare

Agent workflow design
Revise how LLM reasoning, tool calling, retrieval, and workflow orchestration fit together. Exercise: sketch an agent that handles a hypothetical repetitive Commerce Operations request, marking tool decisions and failure paths. These are stated responsibilities; A2A and MCP interoperability are preferred knowledge.
RAG and prompt grounding
Revise how documents, structured data, and enterprise repositories can supply context, and how prompts affect accuracy. Exercise: diagram a retrieval pipeline and identify where irrelevant or outdated evidence could produce a wrong answer. The posting requires working knowledge of RAG and prompt engineering.
Secure integration principles
Revise REST and GraphQL integration, authentication, validation, rate limiting, and error handling. Exercise: review a hypothetical agent-to-API exchange and identify how it should handle unauthorized access, invalid input, and rate limits. Secure enterprise integrations are an explicit responsibility.
Guardrails and human control
Revise where defined guardrails, human approval, LLM routing, and failover belong in an agent workflow. Exercise: mark which actions in your workflow should pause for approval or fail safely. These controls appear in the duties; AI governance and PII handling are preferred knowledge.
Evaluation and operational outcomes
Revise the difference between agent-quality measures and business outcomes. Exercise: define a small test set, a quality-regression check, and measures for accuracy, cycle time, and incidents. The role contributes to evaluation and reports operational results.
Production lifecycle
Revise how version control, CI/CD, monitoring, rollback, and cost tracking support an AI solution after deployment. Exercise: outline what to check before release and what would trigger rollback. The posting explicitly includes deployment and production support.

Skills to revise

Implement a Python agent workflow
Practice building a small tool-calling workflow in Python with one framework named in the posting, such as LangGraph, and test a failed tool response. Python and experience with at least one LLM or agent framework are minimum qualifications.
Build a small RAG pipeline
Practice retrieving from sample documents and structured records, then inspect whether generated answers use the retrieved evidence. Building and maintaining RAG pipelines across those source types is a stated duty.
Handle an API integration
Practice connecting a sample agent workflow to a mock REST or GraphQL endpoint, including authentication, input validation, rate-limit handling, and error responses. The role develops secure integrations with commerce platforms and enterprise data.
Test and trace agent behavior
Practice creating a compact test dataset, recording a trace and audit log for each run, and spotting a quality regression after a prompt change. Evaluation, tracing, audit logging, and troubleshooting are listed responsibilities.
Prepare a controlled release
Practice putting an agent change through Git, tests, and a CI/CD check; write a brief rollback and monitoring plan. Familiarity with these development practices and support for production rollback are explicit in the posting.

Your preparation checklist

Map your evidence to the minimum qualifications
Prepare a one-page inventory of examples you can discuss for Python, an LLM or agent framework, RAG, tool calling, API integration, and Git, testing, debugging, code reviews, and CI/CD. Note your role and contribution in each example; the posting asks for 3+ years in a related field.
Rehearse an AI application walkthrough
Select a project you can discuss and prepare a concise account of the problem, architecture, your implementation, tests, failures, and results. If it used an agent framework, explain why you chose its workflow and where human review or tool calls fit.
Sketch a Commerce Operations agent design
As interview practice, draw a hypothetical workflow for a repetitive task: retrieval from documents or structured data, LLM reasoning, tool calls, and a clear stop or escalation path. Label assumptions rather than presenting the design as a Cisco system.
Prepare a secure integration example
Be ready to explain a REST, GraphQL, webhook, or event-driven integration you have worked on, including authentication, input validation, rate limits, error handling, and troubleshooting. The role explicitly involves secure connections to commerce platforms and enterprise data.
Work through safety and interoperability trade-offs
Prepare a short design exercise showing where defined guardrails, human approval, LLM routing, and failover would apply. If relevant to your experience, be ready to distinguish A2A, MCP, and OpenAPI-based tool integration; the posting lists these protocols as preferred familiarity.
Outline how you would evaluate and operate an agent
Draft a sample test dataset and a monitoring plan covering outcome accuracy, tracing, audit logs, regression detection, rollback, and cost. Explain how you would investigate a failure and report efficiency, cycle-time, or incident metrics without inventing results.
Bring questions for the COEC team
Ask which Commerce Operations tasks are priorities, how senior engineers and business teams define acceptable automated actions, what data and approval constraints apply, and which quality and operational metrics matter most.

Availability last checked: 4 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…