New opportunity

Senior Azure AI Engineer

OSI Digital · Hyderabad, India

About this role

OSI Digital seeks a Senior Azure AI Engineer to deliver production-grade AI solutions on Microsoft Azure from its Hyderabad office. The role covers copilots and agentic workflows, retrieval-augmented search, speech pipelines, and document automation, with attention to business impact, security, reliability, and cost.

Responsibilities

Design and deploy copilots and agents using Copilot Studio and Azure OpenAI, including tool calling, grounding, and guardrails. Build RAG systems with Azure AI Search using indexers, skillsets, vector and hybrid retrieval, semantic ranking, and query rewriting.
Develop batch and real-time speech-to-text, text-to-speech, diarization, and Arabic/English translation pipelines. Automate extraction with Azure AI Document Intelligence using custom and prebuilt models, confidence thresholds, and retries.
Orchestrate workflows through Power Automate and Azure Functions; integrate Microsoft 365 applications and APIs. Establish LLMOps practices covering evaluation sets, A/B tests, prompt versioning, and cost and performance monitoring. Apply Content Safety, PII controls, and Azure security practices. Produce runbooks and architecture documentation and conduct handovers.

Requirements

The job description states 5–7 years of experience; the listing header states 5–8 years. Required skills specify 5+ years in software or AI engineering, including 2+ years building on Azure AI services.
Experience across at least three Azure AI categories—Azure OpenAI, Search, Speech, Document Intelligence, Vision, Translator/Language, or Content Safety—and at least one production deployment. Hands-on experience with Prompt Flow, Semantic Kernel or LangChain, agent patterns, RAG engineering, offline evaluation, and relevance metrics.
Proficiency in Python and/or C#, REST APIs, webhooks, and asynchronous patterns; experience with a vector store such as Azure Cosmos DB, PostgreSQL with pgvector, or Redis. Knowledge of AI workload CI/CD, infrastructure as code, telemetry, and Azure security and governance, including data residency and tenant isolation. Node.js is a plus.

Candidates must be willing to work from the OSI office in Hyderabad, use AI tools such as Cursor or Claude Code, and have worked directly with clients. The role is hands-on, not limited to team management.

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

Skills for this role

Azure OpenAICopilot StudioAzure AI SearchAzure AI SpeechAzure AI Document IntelligenceRetrieval-Augmented GenerationAgent orchestrationFunction callingVector searchSemantic rankingSpeech-to-textText-to-speechDiarizationPower AutomateAzure FunctionsMicrosoft TeamsSharePointOneDriveOutlookPrompt FlowSemantic KernelLangChainLLMOpsAI evaluationA/B testingContent SafetyPII detectionManaged IdentityAzure Key VaultPython

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

Ship grounded copilots and agents
The role calls for hands-on delivery with Copilot Studio and Azure OpenAI, including tool calling, grounding, and guardrails. Prepare to explain production design decisions and how they support business impact.
Engineer and evaluate RAG search
Azure AI Search work includes indexers, skillsets, vector and hybrid retrieval, semantic ranking, and query rewriting. The requirements also emphasize chunking, re-ranking, caching, offline evaluation, and relevance metrics.
Deliver speech and document automation
Responsibilities span batch and real-time speech pipelines, diarization, Arabic/English translation, and Document Intelligence extraction using custom or prebuilt models, confidence thresholds, and retries.
Make AI workloads operable and secure
The posting calls for evaluation sets, A/B tests, prompt versioning, cost and performance monitoring, CI/CD, and telemetry, alongside Content Safety, PII controls, and Azure security and governance.
Integrate workflows and hand them over
The engineer is expected to connect Power Automate or Azure Functions with Microsoft 365 and APIs, then produce architecture documentation, runbooks, and knowledge-transfer materials. Direct client work and hands-on engineering are explicitly expected.

Likely challenges to think through

Balancing retrieval quality against cost and latency
Preparation hypothesis, not a stated employer problem: chunking, hybrid retrieval, semantic ranking, and query rewriting can improve relevance while adding complexity and expense. Prepare to describe how offline relevance metrics and performance monitoring would guide those choices.
Handling uncertain speech and extraction outputs
Preparation hypothesis: real-time speech and document processing may require different latency and accuracy choices, while low-confidence extraction needs a recovery path. Prepare to explain when to use batch processing, confidence thresholds, and retries without treating every result as certain.
Connecting agents to enterprise data safely
Preparation hypothesis: tool calling and Microsoft 365 integrations can broaden access to sensitive or tenant-specific data. Prepare to discuss how grounding, PII controls, identity, private networking, and tenant isolation would shape an integration design.

Practice your answers

Technical practice

How would you design a grounded copilot that calls tools without taking unsafe actions?

What to coverThe role calls for Copilot Studio and Azure OpenAI agents with tool calling, grounding, and guardrails. Suggested answer focus: explain tool permissions, validation of tool outputs, grounding checks, and when to require confirmation before an action.

How would you build an Azure AI Search index and retrieval pipeline for RAG?

What to coverThe posting names indexers, skillsets, chunking, embeddings, hybrid retrieval, semantic ranking, and query rewriting. Suggested answer focus: connect ingestion and chunking choices to retrieval quality, and explain when hybrid search or re-ranking helps.

How would you diagnose a RAG system that returns plausible but irrelevant answers?

What to coverThe role specifies offline evaluation and relevance metrics alongside RAG engineering. Suggested answer focus: separate retrieval failures from generation failures, use an evaluation set to inspect results, and explain how you would test changes to chunking, ranking, or caching.

How would you design batch and real-time speech pipelines for Arabic and English?

What to coverThe posting covers speech-to-text, text-to-speech, diarization, and Arabic/English translation using Azure AI Speech. Suggested answer focus: distinguish latency-sensitive from batch processing, explain where diarization and translation fit, and describe how you would assess output quality.

How would you handle uncertain document-extraction results?

What to coverThe role uses Azure AI Document Intelligence with custom and prebuilt models, confidence thresholds, and retries. Suggested answer focus: discuss model selection, threshold-based handling of low-confidence fields, and how to avoid treating a retry as proof that an extraction is correct.

How would you orchestrate an AI workflow across Microsoft 365 and external APIs?

What to coverThe posting names Power Automate, Azure Functions, Teams, SharePoint, OneDrive, Outlook, REST APIs, webhooks, and asynchronous patterns. Suggested answer focus: explain where each orchestration option fits and how you would handle authentication, delayed responses, retries, and duplicate events.

How would you evaluate and roll out a prompt change in production?

What to coverThe role calls for Prompt Flow, evaluation sets, A/B tests, prompt versioning, and cost and performance monitoring. Suggested answer focus: define quality and operational measures, compare versions on representative cases, and explain how you would detect a regression.

How would you protect sensitive data in a multi-tenant Azure AI solution?

What to coverThe posting specifies Content Safety, PII detection and redaction, Entra ID, Managed Identity, Key Vault, private networking, data residency, and tenant isolation awareness. Suggested answer focus: trace data and access across ingestion, retrieval, model calls, and logs, identifying where each control applies.

How would you deploy and operate an Azure AI workload reliably?

What to coverThe role mentions CI/CD, infrastructure as code, App Insights or OpenTelemetry, and production-grade reliability and cost. Suggested answer focus: describe versioned deployment and rollback, useful telemetry for failures and latency, and how monitoring would inform cost or performance changes.

How would you choose a vector store for a RAG workload?

What to coverThe posting lists Azure Cosmos DB vector search, PostgreSQL with pgvector, and Redis as relevant experience. Suggested answer focus: compare them against an application's existing data architecture, filtering and retrieval needs, latency, and operational cost without claiming one is universally best.

Behavioural practice

Tell me about a time you worked directly with a client to define an AI deliverable.

What to coverDirect client work and measurable business impact are stated aspects of this role. Suggested answer focus: use situation, action, and result to show how you clarified the need, made a technical decision, and assessed its outcome.

Describe a production issue you investigated in an AI or API-backed system.

What to coverThe role emphasizes production reliability, telemetry, and runbooks. Suggested answer focus: give the situation, your diagnostic and recovery actions, and the result, including what you changed to reduce recurrence.

Tell me about a technical handover you made effective for another team.

What to coverArchitecture documentation, runbooks, handover materials, and knowledge-transfer sessions are explicit duties. Suggested answer focus: explain the situation, what you documented or demonstrated, and evidence that the receiving team could operate the solution.

Describe a time you balanced security, quality, cost, and delivery speed.

What to coverThe posting emphasizes all four concerns for production AI solutions. Suggested answer focus: present the situation, the trade-off you made and communicated, and the result; distinguish measured outcomes from expectations.

Tell me about a time you used an AI coding tool while retaining engineering ownership.

What to coverThe posting expects willingness to use tools such as Cursor or Claude Code and describes a hands-on engineering role. Suggested answer focus: use situation, action, and result to explain how you checked generated work, addressed risks, and verified the final change.

Sharpen your knowledge

Topics to prepare

Agent grounding and guardrails
Posting: Copilot Studio and Azure OpenAI agents must use tool calling, grounding, and guardrails. Suggested revision exercise: diagram an agent request from user input through retrieval and tool authorization to a grounded response; identify where unsafe or unsupported output should be blocked.
RAG retrieval and relevance
Posting: The role calls for chunking, embeddings, caching, Azure AI Search hybrid retrieval, semantic ranking, query rewriting, and offline relevance metrics. Suggested revision exercise: compare two chunking and retrieval designs for the same document set, stating which relevance measures would reveal a regression.
Speech and document pipeline decisions
Posting: Work includes batch and real-time speech, diarization, Arabic/English translation, and Document Intelligence extraction with confidence thresholds and retries. Suggested revision exercise: sketch how a pipeline would handle low-confidence fields, failed transcription, and retryable versus non-retryable errors.
LLMOps evaluation
Posting: Responsibilities include Prompt Flow, evaluation sets, A/B tests, prompt versioning, and cost and performance monitoring. Suggested revision exercise: define a small evaluation set and decision criteria for comparing two prompt versions, including quality, latency, and cost.
Azure security and production governance
Posting: The role names Content Safety, PII redaction, Entra ID, Managed Identity, Key Vault, private networking, data residency, and tenant isolation. Suggested revision exercise: review an AI workload architecture and mark identity, secret, network, data-boundary, and PII-control points.

Skills to revise

Build a tool-calling agent
Posting: Hands-on work spans Copilot Studio, Azure OpenAI, and Prompt Flow, Semantic Kernel, or LangChain. Suggested revision exercise: implement a small grounded agent using one listed orchestration option, add a REST-backed tool call, and test both an authorized call and a request the guardrails should reject.
Configure and test Azure AI Search
Posting: RAG implementation includes indexers, skillsets, vector and hybrid retrieval, semantic ranking, and query rewriting; experience with a listed vector store is required. Suggested revision exercise: build a small searchable corpus, compare retrieval modes against labeled queries, and record relevance results.
Prototype a speech workflow
Posting: Azure AI Speech work covers batch and real-time STT, TTS, diarization, and Arabic/English translation. Suggested revision exercise: prototype one batch and one real-time path, then inspect diarization or translation errors and document how each path reports failures.
Handle document extraction uncertainty
Posting: Document Intelligence work uses prebuilt and custom models, confidence thresholds, and retries. Suggested revision exercise: extract fields from sample documents, route low-confidence results for review, and test retry behavior on a failed request.
Orchestrate an M365 integration
Posting: Workflows use Power Automate or Azure Functions and integrate Teams, SharePoint, OneDrive, Outlook, and APIs; Python and/or C#, webhooks, and asynchronous patterns are required. Suggested revision exercise: implement a small event-driven workflow using one listed M365 application, a webhook, and an asynchronous processing step.
Prepare a production release and handover
Posting: The role calls for AI-workload CI/CD, infrastructure as code, telemetry, monitoring, runbooks, architecture documents, and client handovers. Suggested revision exercise: outline a deployment pipeline using GitHub Actions or Azure DevOps, add App Insights or OpenTelemetry signals for errors, latency, and cost, and draft a short incident runbook.

Your preparation checklist

Build a requirements evidence sheet
List actual examples of work across at least three named Azure AI categories, one production deployment, Azure AI service experience, and direct client work. Note the listing’s differing experience ranges: 5–7 years in the description and 5–8 in the header.
Prepare a production project walkthrough
Choose a real deployment you can discuss. Bring a diagram and explain your hands-on contribution, client requirements, business impact, reliability and cost decisions, and what changed after launch.
Practice an agent and RAG design exercise
For a hypothetical client copilot, sketch Copilot Studio or Azure OpenAI tool calling and guardrails alongside Azure AI Search indexing, chunking, vector and hybrid retrieval, semantic ranking, and relevance evaluation. State your trade-offs.
Outline a speech or document pipeline
Prepare a design for a relevant task, such as Arabic/English speech processing or document extraction. Show where diarization or translation, model choice, confidence thresholds, retries, and human review would fit; label it hypothetical unless drawn from your work.
Show an integration and coding example
Select a Python or C# example you can explain, covering REST APIs, webhooks or asynchronous work. Be ready to sketch how Azure Functions or Power Automate could connect the workflow to the Microsoft 365 applications named in the posting.
Prepare an evaluation and operations plan
Draft a sample evaluation set, relevance metrics, an A/B comparison, and a prompt-versioning approach. Explain how you would monitor quality, latency and cost, and deploy and observe changes using the CI/CD and telemetry practices the role requests.
Review security and handover artifacts
Prepare an architecture sketch and a short runbook outline addressing Content Safety, PII handling, identity, secrets, private networking, data residency and tenant isolation. Include what you would document and transfer to an operating team.
Write focused questions for OSI Digital
Ask which Azure AI workloads need attention first, how production quality and business impact are measured, and what client-facing work and Hyderabad office expectations involve. Confirm how hands-on AI coding tools are used in the role.

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…