New opportunity

Software Engineer - AI

LevelShift · Chennai, India

About this role

LevelShift seeks a Software Engineer - AI to develop AI-native applications and intelligent automation that address enterprise business problems. The role works with product teams, software engineers, data professionals and business stakeholders to deliver scalable, secure, production-ready solutions.

Responsibilities

Build LLM pipelines for reasoning, document generation, summarization, content extraction and task automation. Design autonomous workflows in which AI agents use tools, APIs, memory, external systems and other agents, applying prompt engineering, retrieval-augmented generation (RAG) and planning techniques.
Modernize enterprise applications, rule-based systems, stored procedures and legacy workflows into AI-native solutions that process structured and unstructured data. Integrate LLMs with enterprise systems, cloud platforms, databases and business processes while following responsible AI and governance practices.
Identify AI adoption opportunities with internal and customer-facing stakeholders. Use generative AI and AI-powered tools across software development for code generation, prompt optimization, testing, documentation, workflow automation and knowledge retrieval.

Qualifications

Bachelor's degree in Computer Science, Artificial Intelligence, Information Technology, Data Science, Engineering or a related field, with at least 3 years of experience in AI engineering, software engineering, machine learning or generative AI.
Hands-on experience building applications with LLMs, prompt engineering, RAG, embeddings, vector databases and AI agent frameworks such as LangChain, LangGraph, Semantic Kernel, CrewAI or AutoGen. Proficiency in Python, REST APIs, cloud platforms and integrating AI services with enterprise applications, databases and business systems; Azure is preferred.
Understanding of AI governance, responsible AI, model evaluation, security, data processing and modern AI application architectures. Analytical, problem-solving and communication skills and the ability to collaborate in Agile teams are also required. Technologies used include Azure OpenAI, Microsoft Copilot and GitHub Copilot.

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

Skills for this role

Generative AILarge Language Models (LLMs)AI agentsPrompt EngineeringRetrieval-Augmented Generation (RAG)EmbeddingsVector DatabasesLangChainLangGraphSemantic KernelCrewAIAutoGenPythonREST APIsAzureAzure OpenAIMicrosoft CopilotGitHub CopilotAI governanceResponsible AIModel EvaluationSecurityStructured Data ProcessingUnstructured Data ProcessingAgileProblem-SolvingCommunication

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

Build useful LLM pipelines
The role calls for pipelines that handle reasoning, document generation, summarization, content extraction and task automation. Prepare to explain how you would design and evaluate an end-to-end pipeline for a business task.
Develop agentic workflows
The posting emphasizes agents that use tools, APIs, memory, external systems and other agents, alongside prompt engineering, RAG and planning. Focus on when those components add value and how their behavior can be tested.
Modernize enterprise workflows
LevelShift wants existing applications, rule-based systems, stored procedures and legacy workflows analyzed and adapted into AI-native solutions that process structured and unstructured data. Prepare to discuss how existing business logic would be preserved where needed.
Deliver production-ready integrations
The role integrates LLMs with enterprise systems, databases, cloud platforms and business processes. Security, scalability, model evaluation, responsible AI and governance are explicit considerations; Azure is the preferred cloud platform.
Identify adoption opportunities with stakeholders
The engineer works with internal and customer-facing stakeholders to recommend AI solutions and improve processes. The posting also calls for using generative AI tools in development, testing, documentation and knowledge retrieval.

Likely challenges to think through

Balance agent autonomy with reliability
Preparation hypothesis: tool-using agents may automate complex tasks but create more ways for a workflow to fail. Prepare an example of choosing between a bounded pipeline and an agent, including how you would test outputs and tool interactions.
Modernize without losing business rules
Preparation hypothesis: replacing legacy logic with AI may increase flexibility while making established behavior harder to reproduce. Prepare to describe how you would map current rules and data flows, then validate an AI-native approach against them.
Balance delivery speed with governance
Preparation hypothesis: integrating LLMs into enterprise systems may put rapid iteration in tension with security, responsible AI and evaluation. Prepare to explain how you would assess data exposure, output quality and failure cases before production use.
Choose feasible AI opportunities
Preparation hypothesis: stakeholders may see potential in automation before the data, integrations or success criteria are clear. Prepare to show how you would narrow a proposed use case to a measurable business task and surface its dependencies and limits.

Practice your answers

Technical practice

How would you design an LLM pipeline for document summarization and content extraction?

What to coverSuggested answer focus: Explain input preparation, prompts, output structure, and how downstream tasks consume the results. Discuss how you would handle incomplete documents and check whether summaries and extracted fields reflect the source.

How would you improve a prompt that produces inconsistent results on a business task?

What to coverSuggested answer focus: Describe how you would isolate failure cases, revise instructions or examples, and compare outputs against defined criteria. Address the trade-off between more detailed prompts and maintainability.

How would you build retrieval-augmented generation over enterprise documents?

What to coverSuggested answer focus: Cover document preparation, embeddings, vector retrieval, and how retrieved content informs the LLM response. Discuss relevance, access to source material, and what to do when retrieval finds insufficient evidence.

When would you use an AI agent rather than a fixed workflow?

What to coverSuggested answer focus: Compare predictable task steps with work that benefits from planning and tool use. Explain how you would constrain interactions with APIs, memory, external systems, or other agents and handle failed actions.

How would you assess a legacy rule-based workflow for AI-native modernization?

What to coverSuggested answer focus: Identify which rules or stored procedures should remain deterministic and where an LLM could add value. Explain how you would validate the new workflow against existing business behavior before replacing it.

How would you integrate an LLM application with enterprise systems?

What to coverSuggested answer focus: Describe a Python and REST API integration with a database or business application, including data flow and failure handling. Discuss security and scalability trade-offs for a production-ready deployment; Azure is a preferred cloud platform in the posting.

How would you process structured and unstructured data in one AI workflow?

What to coverSuggested answer focus: Explain how you would preserve structured fields while extracting usable information from documents or other unstructured inputs. Address validation and what happens when the two sources conflict.

How would you evaluate an LLM solution before and after release?

What to coverSuggested answer focus: Propose task-specific checks for outputs such as summaries, extracted content, or automated actions. Discuss how representative cases, error analysis, and changes to prompts or retrieval would inform further evaluation.

How would you apply security, responsible AI, and governance to an agent workflow?

What to coverSuggested answer focus: Discuss data access, limits on tool actions, and handling of sensitive information. Explain how you would make outputs or actions reviewable while balancing automation with business risk.

How would you use AI-powered development tools without weakening software quality?

What to coverSuggested answer focus: Draw on the posting’s uses of AI for code generation, testing, documentation, and prompt optimization. Explain how you would review generated work and verify that it meets the task’s requirements.

Behavioural practice

Tell me about a time you identified an opportunity for intelligent automation with a stakeholder.

What to coverSuggested answer focus: Set out the business problem and the stakeholder’s needs, then explain how you assessed whether AI was useful. Describe your action and the resulting decision or outcome without assuming automation was the right answer.

Describe a time you collaborated across product, engineering, data, and business perspectives.

What to coverSuggested answer focus: Give the situation and the differing priorities you needed to reconcile. Explain your contribution, how you communicated trade-offs, and the result.

Tell me about a difficult problem you encountered while developing or integrating an AI application.

What to coverSuggested answer focus: Describe the failure or constraint, how you investigated it, and the options you considered. State what you changed and what the result showed.

Describe a time you changed your approach during an Agile delivery cycle.

What to coverSuggested answer focus: Explain what new information emerged and how you coordinated the change with the team. Cover your actions and the effect on the delivered work.

Tell me about a time you had to explain an AI solution’s limitations to a stakeholder.

What to coverSuggested answer focus: Describe the stakeholder’s goal, the limitation or risk, and how you communicated it. Explain the decision or adjustment that followed and its result.

Sharpen your knowledge

Topics to prepare

LLM pipelines and prompt design
The posting names reasoning, document generation, summarization, content extraction and task automation. As a revision exercise, sketch a pipeline for extracting information from a document, then explain how you would check whether its output follows the requested format.
RAG, embeddings and vector databases
These are explicit hands-on requirements. As a conceptual exercise, trace a question from document processing through embedding, retrieval and answer generation; explain how poor retrieval could affect the answer.
Agent workflow design
The role describes agents using tools, APIs, memory, external systems and other agents. Draw a workflow for a business task and explain when an agent should call a tool, retain context or stop rather than act.
Enterprise modernization architecture
The posting calls for modernizing rule-based systems, stored procedures and legacy workflows while handling structured and unstructured data. Map a hypothetical existing workflow into components, distinguishing rules that should remain deterministic from steps that might use an LLM.
Evaluation, security and responsible AI
Model evaluation, security, governance and responsible AI are stated areas of understanding. Draft an evaluation plan for a document-answering application, covering answer quality, handling of sensitive data and review of unsafe or incorrect outputs.
Identifying business opportunities
The role involves working with internal and customer-facing stakeholders to identify AI adoption opportunities. Practice comparing two hypothetical workflows by business problem, available data, integration needs and how success could be assessed; this comparison is a preparation exercise, not a stated employer process.

Skills to revise

Build and test an LLM pipeline
Practice implementing a small prompt-driven summarization or extraction flow and testing it against varied inputs. This directly reflects the posting’s pipeline responsibilities and prompt-engineering requirement.
Implement a retrieval-backed answer flow
Practice processing sample documents, creating embeddings, retrieving passages from a vector database and checking answers against those passages. The posting explicitly requires RAG, embeddings and vector-database experience.
Prototype a bounded agent workflow
Using one of the listed framework options, such as LangChain, LangGraph or Semantic Kernel, practice an agent that calls an API and records its steps. Review how to limit actions and handle a failed call; agent frameworks and tool-using workflows are named in the posting.
Integrate Python services through REST APIs
Practice writing a Python service that accepts structured input, calls an AI service through a REST API and returns a validated response. Explain where database or enterprise-system integration would fit; Python, REST APIs and those integrations are explicit requirements.
Evaluate and safeguard an AI application
Create a small test set with expected outcomes, record failures and check how the application handles sensitive inputs and unreliable outputs. This exercises the posting’s model-evaluation, security and governance areas without assuming a particular company policy.
Use AI tools in software development
Practice using GitHub Copilot or Microsoft Copilot for a small coding, testing or documentation task, then review and correct the result yourself. The posting names these tools and describes AI-assisted development; the exercise does not assume either tool is mandatory for every task.

Your preparation checklist

Map your evidence to the role
Prepare a one-page summary of your degree, relevant experience, and examples involving Python, REST APIs, cloud platforms, LLMs, RAG, embeddings, vector databases, and agent frameworks. The posting asks for a related degree and 3+ years of relevant experience; note where evidence is unavailable rather than filling gaps.
Rehearse an AI application walkthrough
If you have a relevant project, prepare a short account of its business problem, LLM pipeline, prompt and retrieval choices, integrations, your contribution, and measured results. The role includes summarization, extraction, document generation, and task automation.
Sketch an agent workflow
As a practice exercise, diagram an agent using tools or APIs, memory, and an external system to complete a business task. Explain planning, tool permissions, failure handling, and when a simpler non-agent workflow would suffice.
Practice a legacy-modernization design
Sketch how you would assess a rule-based workflow or stored procedure and connect structured and unstructured data to an AI-native application. Show Python or REST integration boundaries and how you would keep the solution secure and scalable; this is a preparation exercise, not a stated project.
Prepare an evaluation and governance plan
Draft example tests for output quality, retrieval relevance, and failure cases, plus safeguards for sensitive data and responsible AI. Be ready to explain how you would evaluate a production-ready solution, as the posting explicitly mentions model evaluation, security, and governance.
Prepare collaboration examples and questions
Choose an example of working with product, engineering, data, or business stakeholders, including how you identified a useful automation opportunity. Ask LevelShift which business workflows, enterprise integrations, evaluation measures, and AI-assisted development tools this role would prioritize.

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…