New opportunity

Large Language Model Architect

Accenture · Hyderabad, India

About this role

Accenture seeks a Large Language Model Architect in Hyderabad for a hands-on AI LLM Technology Architecture role. The position involves turning architecture into production-quality LLM, generative AI and agentic AI components for client engagements, with a focus on Databricks-based applications, RAG pipelines and platform integrations. The role contributes code, design decisions, documentation and reusable engineering patterns as a technical or engineering lead.

Responsibilities

Build prompts, tools, agents, orchestration flows, memory and context handling, retrieval pipelines and evaluation harnesses. Develop data-grounded applications on the lakehouse using Delta tables, Vector Search, MLflow tracing and evaluation, Model Serving and Unity Catalog governance.
Implement ingestion, parsing, chunking, enrichment, embeddings and retrieval for structured and unstructured enterprise content. Add PII detection and redaction, prompt-injection defenses, content filters, guardrails, access controls, lineage and audit logging.
Work with architects, data engineers, product owners and security stakeholders to deliver tested, observable and maintainable components. Produce component designs, integration specifications, deployment runbooks and evaluation results; account for industry-specific data, process constraints, controls and adoption risks.

Requirements

Minimum 5 years of experience and 15 years of full-time education; a bachelor’s degree in Computer Science, Computer Engineering, Data Science, AI/ML, Information Technology or a related engineering discipline is specified. The qualifications also describe both typically 5+ years of software, data or AI engineering, including 2+ years in cloud-native engineering and 1+ year in GenAI, LLM, NLP or agentic AI delivery, and typically 7+ years of engineering, including 3+ cloud-native years and 1–2+ GenAI-related years.
Hands-on Python, APIs, distributed systems, CI/CD, testing, observability, secure development and delivery of AI/ML or data products in financial services, healthcare, manufacturing, retail, telecom or life sciences. Experience with Databricks AI and lakehouse capabilities, LLM architecture, production AI security and governance, Git, automated testing, infrastructure as code and agile client-facing delivery.

Preferred

Databricks certification; Spark/PySpark, Delta Live Tables, LangChain, LangGraph, LlamaIndex, Haystack, FastAPI, Docker or Kubernetes; model fine-tuning, lakehouse optimization, Responsible AI, model risk management, synthetic data, human-in-the-loop review, A/B testing and 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 AIRAGPythonDatabricksDatabricks Mosaic AIDatabricks Model ServingDatabricks Agent FrameworkMLflowDatabricks Vector SearchUnity CatalogDelta LakeLakehouse MonitoringFeature StoreDatabricks WorkflowsPrompt engineeringEmbeddingsVector databasesAgent orchestrationModel evaluationPII detectionPrompt-injection defensesAPIsDistributed systemsCI/CDAutomated testingObservabilityGitInfrastructure 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-ready Databricks applications
The role turns architecture into working LLM, RAG and agentic AI components. Prioritize how ingestion, parsing, chunking and embeddings connect enterprise content to Delta tables, Vector Search and retrieval.
Evaluate and operate LLM systems
The posting calls for evaluation harnesses, MLflow tracing and evaluation, Model Serving, testing and observability. Prepare to explain how a component moves from a working prototype to a deployed, measurable service.
Engineer security and governance into the design
Explicit duties include PII redaction, prompt-injection defenses, guardrails, authentication, authorization, lineage and audit logging, with governance through Unity Catalog. Treat these as application design concerns, not final checks.
Lead hands-on delivery across stakeholders
This is a coding and technical-lead role working with architects, data engineers, product owners and security stakeholders. It calls for design decisions, integration specifications, deployment runbooks, evaluation results and reusable engineering patterns.

Likely challenges to think through

Retrieval quality versus operating cost
A likely design trade-off is how much content to enrich, index and retrieve before latency and cost become impractical. Prepare a reasoned approach to testing chunking and retrieval choices against answer quality, performance and cost; this is a preparation hypothesis, not a reported Accenture problem.
Agent flexibility versus predictable behavior
Tools, orchestration and memory can make workflows more capable but harder to test and trace. Prepare to describe bounded tool access, evaluation cases and MLflow tracing that would make failures diagnosable. The posting establishes these components, not a specific failure mode.
Useful enterprise data versus access controls
Grounded answers may need content from multiple systems while respecting domain constraints, PII protections and authorization. Prepare to walk through where identity checks, redaction, Unity Catalog governance and audit logging would apply in a retrieval flow; the exact client constraints are not specified.

Practice your answers

Technical practice

How would you turn a high-level LLM architecture into production components?

What to coverThe role involves translating architecture into working LLM applications and platform integrations. Suggested preparation: describe component boundaries, Python APIs, data flows, testing and the decisions you would document before deployment.

How would you build an ingestion and retrieval pipeline for mixed enterprise content?

What to coverThe posting names parsing, chunking, enrichment, embeddings and retrieval for structured and unstructured data. Suggested preparation: explain how content quality and chunking choices affect retrieval, and how you would diagnose missing or irrelevant results.

How would you implement governed RAG on Databricks?

What to coverThe role specifies Delta tables, Vector Search, governed features and Unity Catalog. Suggested preparation: trace a query from authorized data access through retrieval to a grounded response, including how governance applies to retrieved content.

When would you use an agent with tools rather than a simpler LLM workflow?

What to coverThe work includes tool calling, agents, orchestration and memory or context handling. Suggested preparation: compare a fixed workflow with an agentic design, addressing tool permissions, state, failure handling and added complexity.

How would you evaluate a RAG or agentic application before and after release?

What to coverThe posting calls for evaluation harnesses and MLflow tracing and evaluation. Suggested preparation: distinguish retrieval quality from answer quality, explain representative test cases, and show how traces would help investigate regressions.

How would you protect an LLM application handling sensitive enterprise data?

What to coverThe duties explicitly include PII redaction, prompt-injection defenses, content filters, authentication, authorization, lineage and audit logging. Suggested preparation: describe controls at ingestion, retrieval, tool use and output, including what each control cannot guarantee.

What would you check when an LLM service becomes slow or unreliable?

What to coverThe role covers Model Serving, observability, scalability and reliability. Suggested preparation: explain how you would use traces and monitoring to separate retrieval, model invocation and integration failures, then weigh latency against quality.

How would you make LLM components safe to change and deploy?

What to coverThe posting requires Git-based development, automated testing, CI/CD, infrastructure-as-code and secure development practices. Suggested preparation: outline tests for APIs, retrieval and agent behaviour, plus deployment checks and a rollback approach.

How would you manage quality, performance and cost in a lakehouse-based LLM application?

What to coverThe role requires production performance and cost controls; lakehouse optimization and GenAI cost optimization are good-to-have areas. Suggested preparation: discuss how retrieval depth, context size and model invocation affect those trade-offs, and what you would measure before changing them.

How would you decide between improving retrieval and adapting a model?

What to coverThe posting emphasizes RAG and also mentions model training and optional fine-tuning experience. Suggested preparation: compare the approaches in terms of knowledge freshness, evaluation effort, governance and operational cost without assuming fine-tuning is necessary.

Behavioural practice

Tell me about a time you aligned engineering and security stakeholders on a design.

What to coverThe role involves collaboration with architects, data engineers, product owners and security stakeholders. Suggested preparation: use a real situation, your specific actions to resolve differing constraints, and the resulting decision or outcome.

Describe a time you led implementation while contributing hands-on code.

What to coverThe posting describes a hands-on technical or engineering lead who contributes code and design decisions. Suggested preparation: explain the situation, what you personally built or decided, how you coordinated others, and the result.

Give an example of adapting a technical solution to industry-specific constraints.

What to coverThe role calls for attention to domain data, processes, controls and adoption risks in client engagements. Suggested preparation: describe the actual constraint, the design change you made and its observed result; do not claim experience in a domain you have not worked in.

Tell me about a production issue you investigated across system boundaries.

What to coverThe work calls for tested, observable and maintainable components spanning data, models and integrations. Suggested preparation: set out the symptoms, how you narrowed the cause using available evidence, what you changed and what happened afterward.

Describe a time you made an engineering approach reusable for others.

What to coverThe role includes reusable patterns, component designs, integration specifications and deployment runbooks. Suggested preparation: explain the original delivery need, your contribution to the shared artifact or pattern, and evidence that it helped subsequent work.

Sharpen your knowledge

Topics to prepare

LLM and agent architecture
The posting requires RAG, tool calling, agent orchestration, prompt engineering and context handling. Sketch an application that routes a request through retrieval and a tool, then explain where context is retained and how failures are handled.
Lakehouse-grounded retrieval
The role calls for RAG over structured and unstructured enterprise content using Delta tables and Vector Search. Diagram ingestion, parsing, chunking, enrichment, embeddings and retrieval; explain how each stage affects answer grounding.
Databricks application lifecycle
Required experience spans Mosaic AI, Agent Framework, MLflow, Model Serving, Unity Catalog and Databricks Workflows. Map these named capabilities to building, evaluating, deploying, governing and operating one LLM application.
Production AI controls
The posting specifies privacy, security, governance, reliability, scalability and cost controls. Review a hypothetical agent design for PII exposure, prompt injection, excessive access and missing audit records; propose a control for each.
Evaluation and domain constraints
The role requires evaluation metrics and attention to industry data, processes, controls and adoption risks. Define a small evaluation set and success measures for a hypothetical application in one industry named in the posting, clearly stating your assumptions.

Skills to revise

Implement a Python retrieval flow
Python coding and retrieval pipelines are explicit requirements. As a revision exercise, write a small Python flow that chunks sample documents, retrieves relevant passages and passes them as context to a model-invocation interface; add tests for empty and irrelevant results.
Build and inspect an agent workflow
The job calls for prompts, tools, agents and orchestration flows. Draft or implement a two-step agent with one tool call, then test tool failure, unexpected tool output and context limits.
Evaluate and trace an application
MLflow tracing and evaluation, evaluation harnesses and observability are stated responsibilities. Prepare a small set of grounded and unanswerable questions; specify what each trace should capture and how you would compare answer quality and retrieval failures.
Test safety and access boundaries
PII redaction, prompt-injection defenses, authentication, authorization and audit logging are explicit. Write test cases for a sensitive document and a malicious retrieved instruction; specify expected redaction, access denial and audit events.
Prepare production engineering artifacts
The posting calls for Git-based development, automated testing, CI/CD, infrastructure as code and technical documentation. As a preparation exercise, outline a component design, integration specification and deployment runbook for the sample application, including test and rollback checks.

Your preparation checklist

Map your evidence to the stated requirements
Prepare a one-page evidence matrix covering Python, cloud-native engineering, GenAI delivery, Databricks, CI/CD and an industry-domain AI or data product. Note the posting’s minimum of five years and its two different ‘typically’ experience descriptions without treating either as an eligibility verdict.
Prepare a data-grounded application walkthrough
Choose a real project you can discuss, if available, and diagram ingestion, parsing, chunking, embeddings, retrieval and response generation. Identify where Delta Lake, Vector Search, MLflow and Unity Catalog were used; clearly mark any alternatives rather than implying hands-on use.
Practice a Databricks architecture design
As a preparation exercise, sketch an LLM or agentic application with prompts, tools, orchestration, context handling and APIs. Show how it could connect to governed lakehouse data, use Model Serving and handle failures and scale; label the design hypothetical.
Draft a security and governance review
Create a short threat-and-control table for PII, prompt injection, inappropriate content, authentication, authorization, lineage and audit logging. Explain where each control would sit in the application flow.
Prepare evaluation and delivery artifacts
Bring an example or hypothetical evaluation plan with retrieval and answer-quality measures, test cases, MLflow tracing, monitoring and cost checks. Add a brief CI/CD and deployment-runbook outline for a tested, observable component.
Rehearse a technical-lead decision story
Prepare a concise account of a real design decision, code contribution and collaboration with engineering, product or security stakeholders. Include the relevant domain constraints and an artifact you produced; distinguish your own work from the team’s.
Write questions for the employer
Ask which Databricks capabilities the engagement uses today, how evaluation and governance are handled, what artifacts the lead owns, and which domain constraints shape the first client deliverables.

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…