New opportunity

Agentic AI Engineer

INFOSYS LIMITED · Bangalore, India

About this role

Infosys Limited seeks an Agentic AI Engineer in Bangalore for its Engineering Services line. The posting specifies 3–6 years of work experience.

Responsibilities

Design and implement production-grade agentic architectures using large language models and reasoning engines. Build and optimize multi-agent systems for complex task orchestration and collaboration.
Integrate external tools, APIs and knowledge bases to augment agents. Improve agent autonomy, adaptability, reliability and safety, and work with product teams to embed agentic AI in enterprise workflows.
Prototype persistent memory, goal planning and self-correction loops. Contribute to code reviews, technical documentation and team best practices.

Requirements

Strong Python programming and hands-on experience with AI/ML frameworks; working knowledge of LLMs, prompt engineering and basic fine-tuning with LoRA or QLoRA.
Practical experience with agentic frameworks such as LangChain, AutoGen or CrewAI, plus API integration and tool-augmented agent development. Understand responsible AI, safety guardrails and LLM evaluation; be familiar with vector stores, retrieval systems and RAG architectures.
Accepted educational qualifications are a bachelor's degree in engineering, technology, science or computer applications, or a master's degree in technology, engineering, science or computer applications.

Preferred or good to have:

Exposure to MLOps practices, including experiment tracking and model versioning with MLflow or DVC; knowledge of distributed systems and scalable backend architectures; and experience with cloud AI services on Azure, AWS or GCP.
Listed preferred skills include Azure, Google and AWS agentic AI services, Google Cloud Contact Center AI, agent engineering and AgentOps.

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

Skills for this role

PythonAI/ML frameworksLarge language modelsPrompt engineeringLoRAQLoRALangChainAutoGenCrewAIAPI integrationMulti-agent systemsTool-augmented agentsResponsible AIAI safety guardrailsLLM evaluationVector storesRetrieval systemsRAGMLflowDVCDistributed systemsAzure Agentic AI ServicesGoogle Agentic AI ServicesGoogle Cloud Contact Center AIAWS Agentic AI ServicesAgent EngineeringAgentOps

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

Production-grade agent orchestration
The role calls for designing LLM- and reasoning-engine architectures and optimising multi-agent collaboration. Prepare to explain how an agent workflow assigns tasks, coordinates agents and handles failures.
Tool and knowledge integration
The posting requires integrating APIs, external tools and knowledge bases, with familiarity in retrieval and RAG. Prepare a concrete design showing when an agent retrieves information, calls a tool and validates the result.
Reliability, safety and evaluation
Tuning autonomy, adaptability, reliability and safety is an explicit duty; guardrails and LLM evaluation are required knowledge. Prepare to describe measurable checks alongside safeguards for tool use.
Enterprise workflow fit
The engineer is expected to work with product teams to embed agentic AI in enterprise workflows. Prepare to explain how a proposed agent capability would fit an existing process and where human review would remain appropriate.
Iterative engineering
The posting includes prototyping persistent memory, goal planning and self-correction, plus code reviews and documentation. Prepare to distinguish what a prototype demonstrates from what would need further testing before production use.

Likely challenges to think through

Autonomy versus control
A likely design difficulty is allowing agents to plan and act without permitting unsafe or unintended tool calls. Prepare a bounded example with tool permissions, stopping conditions and evaluation of failure cases; this is a preparation hypothesis, not a reported Infosys problem.
Multi-agent complexity versus value
Additional agents may improve task specialisation but make coordination and failure diagnosis harder. Prepare to justify when multiple agents outperform a simpler workflow and how you would inspect handoffs; this trade-off is inferred from the orchestration duty.
Persistent context versus dependable answers
Memory and retrieval may help agents carry context, but stale or irrelevant information can undermine results. Prepare to discuss how you would select, refresh and test retrieved or stored context. This is an inferred challenge based on the listed prototypes and RAG familiarity.

Practice your answers

Technical practice

How would you design a production-grade LLM agent architecture?

What to coverThe posting calls for agentic architectures using LLMs and reasoning engines. Suggested preparation: explain component boundaries, how an agent decides what to do next, and trade-offs between autonomy, reliability and safety.

How would you coordinate agents on a complex task?

What to coverThe role includes multi-agent task orchestration and collaboration. Suggested preparation: describe how you would divide responsibilities, pass context between agents and detect stalled or conflicting work.

How would you make an external API safe and reliable for an agent to use?

What to coverThe posting specifies API integration and tool-augmented agents. Suggested preparation: cover input validation, permissions, error handling and when the agent should stop rather than retry.

How would you structure a Python implementation using an agentic framework?

What to coverStrong Python skills and practical experience with frameworks such as LangChain, AutoGen or CrewAI are listed. Suggested preparation: explain how you would keep agent logic, tool interfaces and tests maintainable without claiming that one framework is required.

How would you connect an agent to a knowledge base using RAG?

What to coverThe posting mentions knowledge bases, vector stores, retrieval systems and RAG. Suggested preparation: walk through retrieval and answer generation, then discuss how you would diagnose irrelevant results or unsupported answers.

When would you change a prompt rather than use LoRA or QLoRA?

What to coverWorking knowledge of prompt engineering and basic fine-tuning is requested. Suggested preparation: compare the effort and evidence needed for each approach, including how you would evaluate whether a change improved the task.

How would you prototype memory, goal planning and self-correction?

What to coverThese three capabilities are explicitly named as prototyping work. Suggested preparation: explain what state an agent should retain, how it would track a goal and how it would recognize and recover from an unsuccessful step.

How would you evaluate an agent's performance?

What to coverThe role calls for LLM evaluation and tuning autonomy, adaptability and reliability. Suggested preparation: propose representative tasks and measures for task completion, erroneous actions and performance across varied inputs.

What guardrails would you put around an autonomous agent?

What to coverResponsible AI, safety guardrails and agent safety are stated requirements. Suggested preparation: address risky tool actions, untrusted retrieved content and conditions for human review or refusal.

What would you check before scaling an agent prototype for enterprise use?

What to coverThe posting calls for production-grade systems; scalable backends, cloud AI services and MLOps are good-to-have areas. Suggested preparation: discuss reliability and operational trade-offs, and distinguish any optional experiment tracking or model versioning from the core requirements.

Behavioural practice

Tell me about a time you worked with a product team to integrate an AI capability into a workflow.

What to coverCollaboration with product teams on enterprise workflows is part of the role. As a suggested practice question, use situation, action and result to show how you clarified workflow needs and responded to feedback.

Describe a time you diagnosed a reliability problem in an AI or software system.

What to coverImproving agent reliability is a stated responsibility. As a suggested practice question, give the situation, your investigation and actions, and the observed result; do not assume the example must involve a multi-agent system.

Tell me about a time you balanced automation against a safety concern.

What to coverThe posting emphasizes both agent autonomy and safety. As a suggested practice question, explain the situation, the trade-off you faced, what you did and the result.

Describe a time you moved a technical prototype toward production use.

What to coverThe role spans prototyping advanced agent capabilities and implementing production-grade architectures. As a suggested practice question, outline the situation, the validation or changes you undertook, and the result.

Tell me about a code review or documentation contribution that improved shared engineering work.

What to coverCode reviews, technical documentation and team best practices are listed duties. As a suggested practice question, provide the situation, your specific contribution and its observable result.

Sharpen your knowledge

Topics to prepare

Agent orchestration and reasoning
The posting calls for production-grade agentic architectures and multi-agent task orchestration. Revision exercise: sketch a two-agent workflow, showing task handoffs, failure paths and when a single agent would be simpler.
Memory, planning and self-correction
The role includes prototyping persistent memory, goal planning and self-correction loops. Revision exercise: diagram what an agent should remember between tasks, how it checks progress and when it should stop retrying.
LLMs, prompting and basic fine-tuning
Working knowledge of LLMs, prompt engineering and basic LoRA/QLoRA fine-tuning is specified. Revision exercise: compare when you would change a prompt, add retrieval or use LoRA/QLoRA for a defined task; explain the trade-offs.
Retrieval and RAG architecture
The posting asks for familiarity with vector stores, retrieval systems and RAG. Revision exercise: trace a question from retrieval through answer generation, identifying where irrelevant results or unsupported answers could arise.
Responsible AI and evaluation
Understanding safety guardrails and LLM evaluation strategies is required; reliability and safety are also responsibilities. Revision exercise: define test cases for unsafe requests, incorrect tool use and unsupported answers, with a pass criterion for each.
Production design and operations
The role emphasizes reliable enterprise workflows; distributed systems, scalable backends and MLOps are good-to-have, not core requirements. Suggested revision exercise: outline failure handling for an agent workflow and, as an optional extension, how MLflow or DVC could track an experiment or model version.

Skills to revise

Python agent implementation
Strong Python and hands-on AI/ML framework experience are required. Suggested exercise: implement a small agent workflow with clear input validation, error handling and tests, then review the code for maintainability.
Agentic framework prototyping
Practical experience with frameworks such as LangChain, AutoGen or CrewAI is specified. Suggested exercise: use one of those named frameworks to prototype two collaborating agents and demonstrate how their tasks and outputs are coordinated.
API and tool integration
The role requires API integration and tool-augmented agents. Suggested exercise: connect a prototype agent to a sample API, validate its inputs and outputs, and handle an unavailable or erroneous response.
Retrieval-backed answers
Vector stores, retrieval and RAG appear in the requirements. Suggested exercise: build a small retrieval-backed question-answering flow over sample documents and inspect which retrieved passages support each answer.
Prompt and evaluation iteration
Prompt engineering, guardrails and LLM evaluation are explicit requirements. Suggested exercise: create a small test set, revise a prompt or guardrail, and record which cases improved or regressed, including a safety case.
Technical review and documentation
The responsibilities include code reviews and technical documentation. Suggested exercise: write a brief design note for the prototype’s agent handoffs, tool permissions and failure behavior, then review the code against that design.

Your preparation checklist

Map evidence to the posting
The role lists 3–6 years of experience, Python and AI/ML frameworks, and specified degree options. Prepare a brief, factual summary of relevant work, education and hands-on technologies; do not claim experience you cannot demonstrate.
Prepare an agent project walkthrough
Choose a project you can discuss in detail. Diagram its LLM and agent architecture, identify your Python contributions and framework choices, and explain tool or API integration, outcomes and trade-offs. These are areas named in the posting.
Practice a multi-agent design exercise
Sketch an enterprise workflow with agent roles, task handoffs and failure handling. Explain where goal planning, persistent memory or self-correction would help; these capabilities are in the responsibilities, while the example workflow is your preparation exercise.
Explain retrieval and model choices
Prepare a small example showing how a knowledge base, vector store and RAG could augment an agent. Be ready to distinguish prompt changes from basic LoRA or QLoRA fine-tuning and explain when you would consider each.
Build an evaluation and safety plan
Draft example tasks, success measures and failure cases for an agent. Include checks for reliability, responsible AI and safety guardrails, all of which the posting calls out.
Document delivery trade-offs
Prepare a short account of code review, technical documentation and collaboration with product teams. If relevant to your experience, add examples of experiment tracking, model versioning, scalable backends or cloud AI services; the posting labels these good to have.
Write targeted employer questions
Ask which enterprise workflows the team is prioritising, how it evaluates agent reliability and safety, and which agentic frameworks or cloud services it uses. These questions are inferred from the advertised work, not stated company processes.

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…