New opportunity

Oliver Wyman – Senior / Lead Data Scientist (AI and Generative AI) - Gurugram

Oliver Wyman Group · Gurugram, India

About this role

Oliver Wyman’s India Data and Analytics practice seeks a hands-on Senior / Lead Data Scientist to design, develop, evaluate and deploy AI and generative AI solutions for global consulting engagements, including work in financial services. The role combines technical delivery with translating client business problems into practical, client-ready solutions.

Responsibilities

Develop AI, machine learning and GenAI solutions using Python, SQL, cloud platforms, LLM APIs, open-source models and AI frameworks. Turn business problems into analytical approaches, prototypes, requirements and scalable implementation plans.
Build and evaluate RAG knowledge assistants, document summarization and extraction tools, workflow copilots, conversational agents, semantic search and prompt-driven analytics. Work with structured data and unstructured sources, including documents, images, transcripts, logs and enterprise knowledge.
Support prompt engineering, embedding and retrieval design, model adaptation, benchmarking and assessments of relevance, accuracy, robustness, hallucination risk, fairness, explainability and business impact. Produce reliable code, APIs, pipelines, demos and reusable assets.
Collaborate with consultants, engineers, designers, governance teams and clients; document findings and limitations, and address privacy, security, data lineage, auditability, compliance and safe deployment.

Required qualifications

3 to 8 years of experience in AI/ML, data science, GenAI engineering, NLP, advanced analytics, software engineering or related consulting or analytics roles, including experience developing ML models, AI applications or data-driven tools.
Bachelor’s or master’s degree in a quantitative or technical discipline. Strong hands-on Python and SQL; knowledge of LLMs, transformers, embeddings, RAG, agents, fine-tuning, evaluation and guardrails; ML fundamentals; unstructured-data preparation or retrieval pipelines; and clear technical and business communication. The role also calls for ownership, attention to detail and managing consulting deadlines.

Preferred experience includes PyTorch, TensorFlow, scikit-learn, Spark, enterprise AI use cases, LangChain, LlamaIndex, Hugging Face, vector databases, MLflow, Databricks, Snowflake, Airflow, Docker, CI/CD and Azure, AWS or Google Cloud. Client-facing consulting, responsible AI governance and support for user testing, training, change management or production rollout are valued. Collaboration across time zones and travel may be required. Marsh’s hybrid arrangement expects at least three days per week in the local office or onsite with clients, including a weekly team anchor day.

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

Skills for this role

PythonSQLMachine LearningData ScienceGenerative AILarge Language ModelsNatural Language ProcessingRetrieval-Augmented GenerationPrompt EngineeringEmbeddingsSemantic SearchFine-TuningModel EvaluationResponsible AIModel GovernanceData PipelinesAPIsTechnical DocumentationPyTorchTensorFlowscikit-learnSparkLangChainLlamaIndexHugging FaceVector DatabasesMLflowDatabricksSnowflakeAirflow

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

Translate client needs into deliverable solutions
Work with partners, consultants and client stakeholders to turn business problems into analytical approaches, prototypes, requirements and scalable implementation plans; communicate findings and limitations in client-ready terms.
Build practical AI and GenAI applications
Develop and evaluate applications such as RAG knowledge assistants, document extraction tools, workflow copilots and semantic search using structured and unstructured data.
Evaluate quality, impact and risk
Benchmark relevance, accuracy, robustness and business impact while assessing hallucination risk, fairness, explainability and human-in-the-loop review.
Deliver reliable, governable assets
Produce clean code, APIs, pipelines, demos and reusable assets, with documentation that addresses privacy, security, data lineage, auditability and safe deployment.

Likely challenges to think through

Prototype speed versus implementation quality
A likely challenge is moving quickly enough for consulting deadlines without leaving a prototype difficult to scale or maintain. Prepare to explain how you would scope an initial demo, then define requirements, tests and a reliable implementation path.
Useful answers versus unsupported outputs
RAG and document tools may return plausible answers that are inaccurate or poorly supported. Prepare an evaluation approach that checks retrieval relevance, answer accuracy and hallucination risk, and identifies when human review is appropriate.
Client value versus governance constraints
Enterprise knowledge can make an application useful while raising privacy, security and auditability concerns. Prepare to discuss how data access, lineage and documented model limitations would shape the design, rather than treating governance as a final check.

Practice your answers

Technical practice

How would you turn an ambiguous client problem into an AI solution plan?

What to coverThe role involves translating business problems into analytical approaches, prototypes, requirements and scalable implementation plans. Suggested preparation: explain how you would define the decision to support, establish a baseline and success measures, then decide whether AI is warranted.

How would you validate a conventional ML model for a client use case?

What to coverThe posting calls for feature engineering, supervised and unsupervised learning, model validation and performance metrics. Suggested preparation: discuss data leakage, appropriate validation splits, baseline comparisons and the trade-off between predictive performance and explainability.

How would you prepare structured and unstructured enterprise data for an AI application?

What to coverThe role covers Python, SQL and preparation or retrieval pipelines for documents, images, transcripts, logs and structured data. Suggested preparation: describe quality checks, transformations, metadata and lineage, including how you would handle missing or inconsistent inputs.

How would you design and troubleshoot a RAG knowledge assistant?

What to coverThe posting explicitly includes RAG assistants, embeddings, retrieval strategy and enterprise knowledge sources. Suggested preparation: walk through ingestion, chunking, indexing, retrieval and grounded responses, then explain how you would isolate retrieval failures from generation failures.

How would you evaluate document summarization or extraction?

What to coverDocument summarization and extraction tools are stated use cases. Suggested preparation: distinguish factual accuracy, completeness and field-level correctness, and explain when human review is needed for consequential errors.

When would you use prompting, RAG or model adaptation?

What to coverThe role includes prompt engineering, retrieval design, fine-tuning or adaptation, and LLM APIs or open-source models. Suggested preparation: compare these approaches by data needs, ability to update knowledge, evaluation results, operational complexity and risk rather than assuming one is always best.

How would you constrain a conversational agent or workflow copilot?

What to coverThe posting names agents, workflow copilots, guardrails and safe deployment. Suggested preparation: discuss boundaries on permitted actions, handling of uncertain outputs and human-in-the-loop review, while distinguishing a prototype from a deployable workflow.

How would you benchmark a GenAI application beyond answer accuracy?

What to coverThe role calls for assessments of relevance, robustness, hallucination risk, fairness, explainability and business impact. Suggested preparation: propose a representative test set, separate component-level from end-to-end measures, and explain how you would examine failures across user groups and use cases.

How would you move a successful prototype toward reliable deployment?

What to coverThe role includes clean code, APIs, pipelines, documentation and reusable assets, with deployment lifecycle knowledge required. Suggested preparation: describe testing, versioning, monitoring and issue handling, and explain trade-offs between speed of delivery and maintainability.

How would you assess privacy and auditability in an AI solution?

What to coverThe posting explicitly includes privacy, security, data lineage, model limitations, compliance and auditability. Suggested preparation: explain what data and output records you would document, how you would surface limitations, and when to involve risk or governance colleagues.

Behavioural practice

Tell me about a time you clarified an ambiguous stakeholder request.

What to coverThe role requires translating client problems into practical solutions with partners, consultants and client stakeholders. Suggested preparation: use a real situation, the questions and decisions you drove, and the resulting scope or outcome.

Describe a time you explained an AI finding or limitation to a nontechnical audience.

What to coverClear, client-ready communication of technical findings, limitations and trade-offs is part of the role. Suggested preparation: give the situation, how you adapted your explanation and recommendation, and what the audience decided or understood as a result.

Tell me about a time you delivered technical work under a tight deadline.

What to coverThe posting emphasizes ownership, attention to detail and managing consulting deadlines. Suggested preparation: describe the situation, your prioritization and quality checks, and the delivery result without glossing over unresolved limitations.

Describe a time you resolved a disagreement across technical and business teams.

What to coverThe role involves collaboration with consultants, engineers, designers, governance teams and client experts. Suggested preparation: identify the competing concerns, the evidence or trade-offs you brought to the discussion, and the resulting decision.

Tell me about a time user feedback changed an analytics or AI solution.

What to coverUser testing, training, change management and rollout support are listed as valued experience. Suggested preparation: use a real situation to show how you gathered feedback, what you changed or chose not to change, and the observed result.

Sharpen your knowledge

Topics to prepare

LLM application design
The posting requires working knowledge of transformers, prompts, RAG, agents, fine-tuning and guardrails. As preparation, compare when a client knowledge assistant should use retrieval, model adaptation or an agent; explain the trade-offs and failure modes.
Retrieval and unstructured data
The role includes embeddings, retrieval strategy and sources such as documents, transcripts and logs. As preparation, sketch how source preparation, indexing and retrieval choices affect answer relevance and traceability.
ML fundamentals and lifecycle
Feature engineering, supervised and unsupervised learning, validation, metrics, experimentation and deployment lifecycle are required knowledge. As preparation, choose a suitable validation plan and metrics for a hypothetical decision-support problem, then explain the risks of data leakage.
Evaluation and responsible AI
The posting calls for assessing accuracy, robustness, hallucination risk, fairness, explainability and business impact, alongside privacy and auditability. As preparation, draft an evaluation plan that pairs quality measures with human review and documents model limitations.
Client problem framing
The role translates business problems into analytical approaches, requirements and scalable plans. As preparation, turn a hypothetical document-intelligence request into a problem statement, success criteria, constraints and a prototype-to-deployment plan.

Skills to revise

Python and SQL delivery
Strong hands-on Python and SQL are required. As preparation, write a small, readable pipeline that joins structured records, prepares text for analysis and checks outputs for missing or duplicate records.
RAG prototyping
Building and evaluating RAG knowledge assistants and semantic search is an explicit responsibility. As preparation, prototype retrieval over a small document set, generate answers from retrieved passages and inspect cases where retrieval fails.
Output benchmarking
The role supports model experimentation, benchmarking and output-quality assessment. As preparation, create a small test set for summarization or extraction, record relevance and accuracy results, and categorize errors and hallucinations.
Reliable interfaces and documentation
The posting asks for reliable code, APIs, pipelines, demos and clear technical documentation. As preparation, package a prototype behind a simple API, add input and error checks, and write a short note on its behavior and limitations.
Client-ready communication
The role requires explaining findings and trade-offs to technical and business audiences. As preparation, present a prototype’s results twice: first as a technical account of methods and limitations, then as a concise recommendation tied to business impact.

Your preparation checklist

Map the requirements to evidence
The posting requires hands-on Python and SQL, ML fundamentals, GenAI knowledge, unstructured-data pipelines and clear communication. Prepare a one-page evidence map linking each area to work you can discuss; mark any gaps rather than inventing examples.
Prepare a client-ready project walkthrough
Choose an AI or analytics project you can discuss. Rehearse the business problem, your contribution, data and technical choices, prototype-to-delivery path, measured results and limitations. The role emphasizes hands-on delivery and practical business impact.
Practice a GenAI design exercise
Sketch how you would turn a client knowledge-assistant or document-extraction request into requirements, a prototype and an implementation plan. Show data preparation, retrieval or prompting choices, an API or pipeline boundary, and trade-offs; this is preparation, not a stated interview exercise.
Build an evaluation and risk plan
Prepare a sample test set and scorecard covering relevance, accuracy, robustness, hallucination risk and business impact. Explain how human review, privacy, security, lineage and auditability would affect a deployment decision; these are concerns named in the posting.
Rehearse technical and business explanations
Prepare short explanations of Python/SQL implementation decisions, ML validation and GenAI concepts such as embeddings, RAG, fine-tuning and guardrails. Practice presenting findings and limitations once to an engineer and once to a client stakeholder.
Prepare delivery and collaboration examples
Select examples you can substantiate of owning work against deadlines, documenting decisions or coordinating with other functions. The role calls for collaboration with consultants, engineers, governance teams and clients across global engagements.
Write focused questions for the employer
Ask which use cases need attention first, how solution quality and business impact are measured, who owns governance and production rollout, and how the senior and lead scopes differ. The posting describes these activities but does not specify their team-level arrangements.

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…