New opportunity

Large Language Model Architect

Accenture · Coimbatore, India

About this role

Accenture seeks a Large Language Model Architect in Coimbatore for a hands-on AI LLM technology architecture role. The position designs, builds, integrates, tests and operationalizes enterprise LLM, generative AI and agentic AI components, with particular responsibility for production-quality applications and integrations on Snowflake. The engineer contributes code, technical decisions, reusable patterns and documentation as a technical or engineering lead.

Responsibilities

Build prompts, tools, agents, orchestration flows, context handling, retrieval pipelines and evaluation harnesses for LLM applications, RAG systems and multi-agent workflows.
Develop Cortex-powered applications and agents within Snowflake’s security perimeter. Use Cortex Search and Cortex AI Functions for RAG and document intelligence, integrate Streamlit and Snowpark applications, and apply access controls, masking, lineage and monitoring.
Implement ingestion, parsing, chunking, enrichment, embeddings, vector search and retrieval for structured and unstructured enterprise content. Add PII detection and redaction, prompt-injection defenses, content filters, guardrails, authentication, authorization and audit logging.
Work with architects, data engineers, product owners and security stakeholders; maintain component designs, integration specifications, deployment runbooks, evaluation results and reusable engineering patterns. Account for domain-specific data, controls and adoption risks in industries such as banking, insurance, retail, healthcare, travel, logistics or telecom.

Requirements

A bachelor’s degree in a related engineering or technology discipline and 15 years of full-time education are specified. The posting requires a minimum of 5 years’ experience and describes typical paths of 5+ years in software, data or AI engineering, including 2+ years in cloud-native engineering and 1+ year in GenAI, LLM, NLP or agentic AI delivery; or 7+ years of engineering experience, including 3+ years in cloud-native architecture or engineering and 1–2+ years in GenAI, LLM, NLP or agentic AI delivery. Candidates need hands-on Python, Snowflake Cortex and related Snowflake tools; LLM architecture, RAG, agents and evaluation; secure software delivery, APIs, CI/CD and observability; and experience delivering AI/ML or data products in at least one listed industry.

Preferred

SnowPro certification or further Snowflake ML exposure; Snowpark Python, semantic models, dbt, Native Apps, governance and performance tuning; exposure to LangChain, LangGraph, LlamaIndex, Haystack, MLflow, FastAPI, Docker or Kubernetes; and experience with Responsible AI, model risk, synthetic data, human-in-the-loop review, A/B testing or GenAI cost optimization.

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

Skills for this role

Large Language Models (LLMs)Generative AIAgentic AIRAGPythonSnowflakeSnowflake Cortex AICortex AgentsCortex SearchCortex AnalystCortex AI FunctionsSnowparkStreamlitSnowflake MLDynamic TablesSnowflake TasksSnowflake StreamsRBACPrompt EngineeringEmbeddingsVector SearchVector DatabasesAgent OrchestrationLLM EvaluationAPI DevelopmentDistributed SystemsCI/CDAutomated TestingObservabilityInfrastructure as Code

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 production LLM applications
The role designs, codes, tests and operationalizes prompts, tools, agents, orchestration, context handling and evaluation harnesses for LLM applications, RAG and multi-agent workflows.
Deliver within Snowflake
The posting emphasizes Cortex-powered applications inside Snowflake’s security perimeter, including Cortex Search and AI Functions, with Snowpark and Streamlit integrations, RBAC, masking, lineage and monitoring.
Engineer enterprise retrieval and controls
Responsibilities span ingestion through parsing, chunking, embeddings and retrieval for structured and unstructured content, alongside PII redaction, prompt-injection defenses, authorization and audit logging.
Lead maintainable client delivery
This is a hands-on technical or engineering lead role: contribute code and design decisions, work with product, data and security stakeholders, and maintain specifications, runbooks, evaluation results and reusable patterns informed by industry constraints.

Likely challenges to think through

Balance retrieval quality with operating cost
A likely design trade-off, not a stated employer problem: richer parsing, chunking and retrieval may improve answers while increasing latency and cost. Prepare to explain how you would compare retrieval and answer-quality evaluations against performance and cost measures.
Keep agents useful within security boundaries
A likely challenge is balancing tool access and agent autonomy against PII protection, authorization and auditability. Prepare a concrete design rationale for permission checks, guardrails, logging and tests inside the stated Snowflake environment.
Turn designs into operable components
A likely difficulty is reconciling rapid iteration with reliable deployment across client-facing integrations. Prepare to describe how you would use automated tests, CI/CD, observability, runbooks and evaluation results to guide changes and communicate trade-offs to stakeholders.

Practice your answers

Technical practice

How would you turn an enterprise LLM use case into a production application architecture?

What to coverThe role covers model invocation, prompts, tools, retrieval and integrations. Suggested preparation: explain how you would define component boundaries and choose between a direct LLM call, RAG and an agent, including the complexity each adds.

How would you prepare structured and unstructured content for retrieval?

What to coverThe posting names ingestion, parsing, chunking, enrichment, embeddings and vector search. Suggested preparation: discuss how content structure and update frequency affect those choices, and how you would check that retrieved passages remain useful and traceable.

How would you build a RAG application within Snowflake’s security perimeter?

What to coverThe role specifically names Cortex Search, Cortex AI Functions, Snowpark and Streamlit in Snowflake. Suggested preparation: describe how these components could support retrieval, generation and an application interface while preserving governed access to enterprise data.

When would you use Cortex Analyst rather than document retrieval?

What to coverCortex Analyst is listed among the required tools, alongside Cortex Search. Suggested preparation: compare questions over structured data with questions answered from documents, and explain how you would validate that the chosen route matches the user’s intent.

How would you design a tool-using or multi-agent workflow?

What to coverThe role calls for tools, agents, orchestration flows and memory or context handling. Suggested preparation: explain tool permissions, context limits, failure handling and when a simpler workflow would be preferable.

How would you evaluate an LLM application before and after deployment?

What to coverEvaluation harnesses, evaluation metrics, monitoring and continuous improvement appear in the posting. Suggested preparation: distinguish retrieval quality, answer quality, safety and operational measures, and describe how test cases and production feedback would reveal regressions.

How would you defend an enterprise LLM application against data exposure and prompt injection?

What to coverThe posting explicitly calls for PII redaction, prompt-injection defenses, guardrails, RBAC, masking, authentication, authorization and audit logging. Suggested preparation: explain where each control belongs and how you would test that retrieved content or tool output cannot bypass access rules.

How would you operationalize Snowflake-based AI components as enterprise content changes?

What to coverThe role lists Dynamic Tables, Tasks, Streams, lineage and observability. Suggested preparation: discuss how you would keep ingestion and retrieval data current, monitor failures and trace outputs back to their source data.

How would you deliver and troubleshoot a production LLM integration?

What to coverPython, APIs, distributed systems, automated testing, CI/CD and secure software delivery are required. Suggested preparation: describe integration tests, deployment checks and diagnostic signals you would use to isolate an application, retrieval or model-invocation failure.

How would you balance quality, reliability and cost in an LLM solution?

What to coverThe posting calls for performance, scalability, reliability and cost controls; GenAI cost optimization is preferred. Suggested preparation: compare options such as retrieval scope, context size and model invocation frequency, and explain what measurements would justify a change.

Behavioural practice

Tell me about a time you led a technical design into working software.

What to coverThe role is a hands-on technical or engineering lead contributing code and design decisions. Suggested preparation: use a situation, your specific actions and the result to show how you balanced architecture with implementation.

Describe a time you resolved conflicting priorities across technical and business stakeholders.

What to coverThe posting calls for collaboration with architects, data engineers, product owners and security stakeholders. Suggested preparation: give a real situation, how you reached a decision or trade-off, and the resulting outcome.

Tell me about a time industry-specific data or controls changed your solution design.

What to coverThe role asks for attention to domain data, process constraints, controls and adoption risks in a listed industry. Suggested preparation: describe the situation, what you changed and the result without claiming experience in an industry you have not worked in.

Describe a time testing or monitoring exposed a problem after an initial design decision.

What to coverThe role emphasizes tested, observable, maintainable components and continuous improvement. Suggested preparation: explain the situation, your diagnosis and corrective action, and the measurable or observable result.

Tell me about a time documentation or a reusable pattern helped another team.

What to coverThe posting names component designs, integration specifications, runbooks and reusable engineering patterns. Suggested preparation: identify the handoff problem, what you created or improved, and how you determined whether it helped.

Sharpen your knowledge

Topics to prepare

LLM, RAG and agent architecture
The posting requires RAG, tool calling, agent orchestration, context handling and model invocation. Suggested revision: draw an architecture for a document-answering agent and explain when it should retrieve, call a tool, retain context or stop.
Snowflake application architecture
The role centers on Cortex-powered applications inside Snowflake’s security perimeter. Suggested revision: trace how Cortex Search, Cortex AI Functions, Cortex Agents, Snowpark and Streamlit in Snowflake could fit together, identifying where data access is enforced.
Security, privacy and governance
The posting calls for RBAC, masking, PII redaction, prompt-injection defenses, lineage and audit logging. Suggested revision: threat-model a RAG request from ingestion through response and identify a control and verification step at each boundary.
Evaluation and production trade-offs
Evaluation metrics, observability, reliability, scalability and cost controls are required. Suggested revision: define a small evaluation plan for answer quality, retrieval quality, safety, latency and cost, then explain what each measure would reveal.
Industry constraints and technical leadership
The role applies domain data and controls in a listed industry and calls for design decisions and engineering documentation. Suggested revision: choose one listed industry and outline its likely data constraints, then write a brief design decision that distinguishes stated job expectations from your proposed solution.

Skills to revise

Build an ingestion and retrieval pipeline
The posting specifies Python and ingestion, parsing, chunking, enrichment, embeddings and vector search. Suggested exercise: implement a small Python pipeline over sample documents, retrieve passages for test questions and inspect failures caused by chunk boundaries.
Prototype a Cortex-backed application
The posting requires hands-on Cortex Search, Cortex AI Functions, Snowpark and Streamlit in Snowflake. Suggested exercise: in an available Snowflake test environment, connect retrieval and an AI function to a minimal Snowpark or Streamlit interface; otherwise, draft the integration and test plan without claiming to have run it.
Implement a controlled agent flow
The role builds prompts, tools, agents and orchestration flows. Suggested exercise: code a small Python agent with one bounded tool, explicit context handling and a stop condition; test tool selection and failure cases.
Test access and safety controls
The posting requires authentication, authorization, masking, PII protection, guardrails and audit logging. Suggested exercise: create test cases for unauthorized retrieval, sensitive-text exposure and prompt injection, and specify the expected block, redaction or logged event.
Ship a tested, observable component
Git-based development, automated testing, CI/CD, observability and deployment runbooks are required. Suggested exercise: add unit and integration tests plus structured operational signals to one of the prototypes, then draft a short deployment and rollback runbook.

Your preparation checklist

Map the stated requirements
Prepare a one-page evidence map for Python, Snowflake Cortex, LLM delivery, cloud-native engineering, secure delivery and experience in a listed industry. Mark any gaps rather than implying experience you do not have.
Rehearse a production project walkthrough
Select a real AI/ML or data product, if available, and explain your hands-on contributions, architecture decisions, integrations, tests and operational outcomes. Distinguish your work from the team’s.
Sketch a Snowflake-based RAG design
As interview preparation, draw a design for ingesting and chunking enterprise documents, creating embeddings, retrieving with Cortex Search and using Cortex AI Functions. Show where Snowpark or Streamlit would fit; label the design as hypothetical.
Prepare an agent and evaluation exercise
Outline prompts, tool calls, orchestration and context handling for a sample workflow. Bring a proposed evaluation table covering answer quality, retrieval relevance, safety, latency and cost; these are preparation choices, not stated interview tasks.
Explain security and governance decisions
Annotate the design with RBAC, masking, PII redaction, prompt-injection defenses, authentication and audit logging. Prepare to explain trade-offs for regulated enterprise content.
Collect delivery artifacts
Bring or describe permissible examples of Python code, API contracts, automated tests, CI/CD, monitoring, component designs and runbooks. Use redacted examples where client material cannot be shared.
Draft questions for the employer
Ask which Snowflake Cortex components the team uses, how it evaluates and monitors production LLM applications, and how architects, data engineers, product owners and security stakeholders divide design and delivery responsibilities.

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…