New opportunity

Principal Architect AI Data Engineer

EXL Talent Acquisition Team · Gurugram, India

About this role

Lead enterprise GenAI and agentic AI architecture for EXL Talent Acquisition Team in Gurugram, Haryana, India. The role combines hands-on technical oversight with organisation-wide architecture, governance, capability building and client-facing leadership.

Responsibilities

Design single-agent, multi-agent and tool-driven systems; establish reference architectures, reusable frameworks and best practices for LLM applications. Oversee end-to-end RAG pipelines spanning ingestion, chunking, embeddings, vector search, orchestration and response synthesis, while improving scalability, reliability, cost and performance.
Guide prompt engineering, agent workflows, tool and function calling, and memory handling. Lead hallucination reduction, guardrails, safety mechanisms and output evaluation.
Architect production-grade APIs and services for LLM solutions. Define MLOps and LLMOps practices for CI/CD, monitoring, observability and evaluation. Partner with data engineering teams on data quality, lineage, governance, compliance and enterprise-platform integration.
Build a GenAI and agentic AI Centre of Excellence, develop reusable components and tooling standards, and support organisation-wide adoption. Work with executives, business stakeholders and clients on AI solutions, pre-sales, RFPs and workshops; influence AI strategy and roadmaps. Establish responsible-AI policies covering security, privacy, data access, redaction, model use, auditability and explainability. Mentor technical teams and review architecture quality.

Required background:

15+ years of total experience in data engineering, data science or AI, including 3+ years of hands-on LLM or GenAI solutions at scale; proven architecture, solution-design and enterprise-delivery experience.
Hands-on experience with LLMs, RAG and retrieval optimisation, GPT and agentic AI implementations, and frameworks such as LangChain or LangGraph. Strong Python and PySpark engineering, API integration, large-volume data analysis, and data engineering integration with Fabric, Azure Databricks or Snowflake are required. Exposure to Azure, AWS or GCP, SQL, containers, CI/CD and monitoring is sought. Prior experience in at least one of data engineering, data science or the ML lifecycle is mandatory; the posting also identifies analytics engineering and data products among its foundations.

Preferred

Fine-tuning approaches including LoRA, PEFT, prompt tuning and few-shot learning; enterprise GenAI deployment and governance; Azure OpenAI, AI Search or Fabric; and use cases in insurance, BFSI, healthcare or retail.

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

Skills for this role

Generative AIAgentic AILLMsRAGVector searchEmbeddingsPrompt engineeringLangChainLangGraphAgent orchestrationTool callingLLM evaluationGuardrailsPythonPySparkFastAPIFlaskAPI integrationMicroservicesMLOpsLLMOpsCI/CDMonitoringData engineeringData governanceData lineageETLELTData scienceNLP .eol

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

Enterprise GenAI architecture
The posting calls for leading single-agent, multi-agent and tool-driven designs, defining reference architectures and reusable frameworks, and overseeing RAG from ingestion through response synthesis.
Production engineering and data integration
Architect LLM APIs and services, define CI/CD, monitoring, observability and evaluation practices, and partner with data engineering on quality, lineage and enterprise-platform integration.
Evaluation and responsible-AI governance
Lead hallucination reduction, guardrails and output evaluation while establishing policies for security, privacy, data access, redaction, model use, auditability and explainability.
Organisation-wide capability building
Build the GenAI and agentic AI Centre of Excellence, standardise reusable components and tooling, mentor technical teams, and review architecture quality.
Executive and client-facing solution leadership
Translate business problems into AI solutions with executives and clients; lead pre-sales, RFP responses and workshops, and influence roadmaps and investment decisions.

Likely challenges to think through

Balancing agent complexity against operational cost
Preparation hypothesis: More elaborate agent workflows may improve capability while increasing latency, cost and failure paths. Prepare to explain when a simpler design suffices and how to compare options on performance, reliability and cost.
Making retrieval dependable at enterprise scale
Preparation hypothesis: Changes in data quality, chunking and retrieval can undermine answer quality even when the model performs well. Prepare an end-to-end RAG evaluation approach that connects retrieval results, response quality and monitoring.
Reconciling reuse with varied client needs
Preparation hypothesis: Organisation-wide frameworks may accelerate delivery but constrain use-case-specific designs. Prepare to describe which architecture elements should be standardised, where variation is warranted, and how design reviews would handle exceptions.
Enabling useful AI within governance boundaries
Preparation hypothesis: Data access and privacy controls may limit what agents can retrieve or retain. Prepare to discuss how redaction, permissions, auditability and evaluation can be designed into workflows without losing sight of the business objective.

Practice your answers

Technical practice

How would you choose between a single-agent, multi-agent, and tool-driven architecture for an enterprise LLM application?

What to coverThe posting calls for all three architectures and for reusable reference designs. Preparation suggestion: Explain how task complexity, control, reliability, latency, and cost affect the choice, including when a simpler design is preferable.

How would you design an end-to-end RAG pipeline?

What to coverThe posting specifies ingestion, chunking, embeddings, vector search, orchestration, and response synthesis. Preparation suggestion: Walk through the data flow and explain how you would maintain source quality and trace a poor answer back to a pipeline stage.

How would you diagnose and improve weak retrieval in a RAG application?

What to coverRAG retrieval optimisation is an explicit requirement. Preparation suggestion: Distinguish retrieval failures from generation failures, then discuss how you would evaluate changes to chunking, embeddings, or search against relevance and cost.

How would you structure prompts, tool calls, and memory in an agent workflow?

What to coverThe role guides prompt orchestration, tool and function calling, memory handling, and frameworks such as LangChain or LangGraph. Preparation suggestion: Explain how you would constrain tool use, manage state, and handle failed calls without adding unnecessary agent complexity.

How would you evaluate LLM outputs and reduce hallucinations?

What to coverThe posting calls for output evaluation, guardrails, safety mechanisms, and hallucination reduction. Preparation suggestion: Describe what you would measure, how you would assemble representative test cases, and the trade-off between stricter controls and useful responses.

What would a production-grade API for an LLM solution need?

What to coverThe role includes architecting APIs and services using FastAPI, Flask, or enterprise microservices. Preparation suggestion: Cover request handling, integration boundaries, failure behaviour, and how you would assess latency, reliability, and scalability.

How would you define CI/CD, monitoring, and evaluation for an LLM service?

What to coverDefining MLOps and LLMOps pipelines, observability, and evaluation is a stated responsibility. Preparation suggestion: Explain what you would test before release, observe after deployment, and use to decide whether a model or prompt change should remain in production.

How would you integrate large-volume enterprise data into a GenAI platform?

What to coverThe posting seeks Python and PySpark engineering, large-volume data handling, and integration with Fabric, Azure Databricks, or Snowflake. Preparation suggestion: Describe how pipeline design, data quality, lineage, and governance would affect downstream LLM use.

How would you turn responsible-AI principles into enforceable architecture controls?

What to coverThe role calls for policies covering security, privacy, data access, redaction, model use, auditability, and explainability. Preparation suggestion: Connect those policies to concrete control points across data ingestion, retrieval, model calls, and outputs, and explain how you would check that controls work.

When would you consider fine-tuning rather than prompt-based approaches?

What to coverLoRA, PEFT, prompt tuning, and few-shot learning are listed as preferred, not required. Preparation suggestion: Compare their likely data, evaluation, maintenance, and cost implications before recommending an approach.

Behavioural practice

Tell me about a time you established reusable technical standards across teams.

What to coverThe posting calls for a GenAI Centre of Excellence, reusable components, and organisation-wide adoption. Preparation suggestion: Use a real situation, explain the actions you took to gain adoption, and describe the resulting change.

Tell me about a time you translated an executive or business problem into a technical roadmap.

What to coverThe role involves engaging executives and business stakeholders and influencing AI strategy and investment decisions. Preparation suggestion: Describe the situation, how you clarified priorities and trade-offs, and what resulted from your recommendation.

Tell me about a time you shaped a client-facing solution under uncertainty.

What to coverClient workshops, solutioning, pre-sales, and RFP responses are stated responsibilities. Preparation suggestion: Give a real situation, your actions to establish scope and communicate limitations, and the outcome without overstating what the solution could do.

Tell me about a time you improved a team's architecture quality.

What to coverThe posting calls for mentoring architects, engineers, and data scientists and reviewing solution designs. Preparation suggestion: Describe the situation, the feedback or coaching you provided, and the observable result.

Tell me about a time you resolved a conflict between delivery goals and governance requirements.

What to coverThe role combines enterprise delivery with security, privacy, compliance, and data-governance responsibilities. Preparation suggestion: Use a real situation, explain how you worked with the affected teams and made the trade-off, and state the result.

Sharpen your knowledge

Topics to prepare

Enterprise GenAI architecture
The posting calls for reference architectures for single-agent, multi-agent and tool-driven systems, with attention to scale, reliability, cost and performance. Suggested revision exercise: sketch two architectures for the same business problem and explain when the added complexity of multiple agents is justified.
RAG design and retrieval optimisation
The role covers ingestion, chunking, embeddings, vector search, orchestration and response synthesis. Suggested revision exercise: trace a query through those stages and explain how you would diagnose poor retrieval separately from poor answer synthesis.
LLM limitations and evaluation
The posting explicitly names hallucination reduction, guardrails, safety and output evaluation. Suggested revision exercise: define a small evaluation set and describe how you would assess grounded answers, unsafe outputs and failures after a model or prompt change.
LLMOps and data governance
The role includes CI/CD, monitoring and observability, alongside data quality, lineage, access, redaction and auditability. Suggested revision exercise: map the controls and checks you would place from data ingestion through production monitoring.
Centre of Excellence and architecture leadership
Building a GenAI Centre of Excellence, setting reusable standards, reviewing designs and influencing strategy are stated responsibilities. Suggested revision exercise: outline a reference-architecture review checklist and explain which decisions should be standardised versus left to delivery teams.

Skills to revise

Agent workflow implementation
The posting seeks experience with LangChain, LangGraph or similar frameworks, plus tool calling, memory and agent orchestration. Suggested revision exercise: build a small tool-driven workflow and test its behaviour when a tool fails or returns ambiguous results.
Python and PySpark data integration
Production-grade Python and PySpark, large-volume data handling, and integration with Fabric, Azure Databricks or Snowflake are named requirements. Suggested revision exercise: write a small PySpark ingestion and quality-check flow, then diagram how its curated output would feed a RAG pipeline.
Production LLM API design
The role calls for production-grade LLM APIs and services, naming FastAPI, Flask and enterprise microservices. Suggested revision exercise: specify an endpoint’s request and response, error handling, access controls and observability, then implement a minimal version in FastAPI or Flask.
Retrieval and response testing
RAG retrieval optimisation and output evaluation are explicit in the posting. Suggested revision exercise: compare two chunking or retrieval configurations on the same test queries, record retrieval and answer-quality results, and explain the trade-offs.
Client-facing solution communication
The architect is expected to work with executives and clients on solutioning, RFPs and workshops. Suggested revision exercise: prepare a short solution brief that states a business problem, proposed architecture, governance controls, delivery trade-offs and open decisions.

Your preparation checklist

Map evidence to the requirements
Prepare a one-page evidence map for the stated 15+ years in data engineering, data science or AI, 3+ years of hands-on LLM/GenAI work at scale, and enterprise architecture and delivery. Include only experience you can substantiate.
Prepare an end-to-end RAG walkthrough
If you have a relevant project, outline ingestion, chunking, embeddings, vector search, orchestration and response synthesis. Note your design decisions and measured effects on retrieval quality, reliability, latency or cost.
Practise an agent architecture design
Sketch a reference architecture for a hypothetical enterprise task. Explain when you would use a single agent, multiple agents or tool-driven workflows, including tool calling, memory handling and failure controls.
Show production engineering decisions
Prepare examples or design notes covering Python/PySpark, API integration, large-volume data handling and integration with Fabric, Azure Databricks or Snowflake. Explain how you would approach CI/CD, monitoring and observability for an LLM service.
Draft an evaluation and governance plan
Create a short test plan for hallucinations, output quality and guardrails, alongside proposed controls for data access, redaction, privacy, model use, auditability and explainability. Present it as your approach, not an existing EXL policy.
Outline a Centre of Excellence approach
Prepare a proposed set of reusable components, architecture review standards and upskilling activities. If you have led comparable work, bring specific examples of adoption or delivery outcomes.
Rehearse a client-facing solution discussion
Use a hypothetical business problem to practise framing an AI solution for executives or clients: scope, alternatives, risks, delivery phases and investment trade-offs. The posting includes workshops, pre-sales and RFP work.
Write questions for the employer
Ask which GenAI platforms and data integrations are in scope, how the Centre of Excellence will work with delivery teams, and what success measures and governance priorities apply to this role.

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…