New opportunity

Data Scientist, Specialist

Vanguard · Hyderabad, India

About this role

Vanguard seeks a Data Scientist, Specialist to lead the design, governance, and deployment of enterprise AI agents at its Hyderabad, Telangana office. The role combines LLMs, enterprise data, and agentic workflows to automate knowledge-intensive processes and deliver measurable business outcomes in partnership with business, risk, compliance, and technology teams.

Responsibilities

Build knowledge systems that integrate structured and unstructured data across business platforms so agents can retrieve, reason over, and use trusted organizational knowledge.
Identify and prioritize business problems suited to agentic AI, LLMs, and NLP. Architect agents that plan, reason, and act across tools, workflows, and data sources, including multi-agent and tool-augmented LLM solutions for complex processes.
Define evaluation frameworks covering coverage, relevance, trustworthiness, explainability, user adoption, and model performance. Work with engineers to deploy scalable, secure, reliable solutions; monitor bias, drift, and failures, implement guardrails, and align systems with risk, compliance, and responsible AI standards.
Advise stakeholders on appropriate AI applications, develop reusable practices and patterns, and mentor teammates.

Qualifications

Minimum 8 years in data science, predictive analytics, or advanced statistical modeling; 3+ years delivering AI/ML solutions in production; and 5+ years of hands-on Python experience.
Experience designing agents and building enterprise LLM applications involving RAG, tool use, orchestration, and evaluation; strong NLP experience with unstructured text and language-driven workflows.
The posting specifies a B.E./B.Tech in Analytics, Applied Mathematics, Economics, Statistics, or Computer Science, or a master's degree/diploma in a related quantitative field. It also lists an MS or PhD in Computer Science, Machine Learning, Data Science, or a related quantitative field.
Strong business problem-solving, communication, and cross-functional enterprise collaboration skills. Distributed data processing experience is a plus.

Only qualified external applicants will be considered. Vanguard describes a hybrid model for most employees and offers financial, physical, and personal wellness benefits.

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

Skills for this role

AI agentsAgentic AILarge Language Models (LLMs)Retrieval-Augmented Generation (RAG)Natural Language Processing (NLP)PythonData sciencePredictive analyticsStatistical modelingEnterprise knowledge systemsMulti-agent systemsLLM orchestrationTool useAI evaluationAI monitoringResponsible AIRisk managementStakeholder collaborationCommunication

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

Choose problems with measurable value
The posting calls for partnering with business stakeholders to identify and prioritize high-value uses of agentic AI, LLMs, and NLP, with a focus on business outcomes rather than model performance alone.
Build trusted enterprise knowledge systems
The role designs systems that combine structured and unstructured data across business platforms so agents can retrieve and use organizational knowledge.
Architect agents for multi-step work
The role leads development of agents that reason, plan, and act across tools, workflows, and data sources, including multi-agent and tool-augmented LLM solutions.
Evaluate and govern production systems
The posting calls for business-centric evaluation of coverage, relevance, trustworthiness, explainability, and adoption, alongside secure deployment, monitoring, guardrails, and alignment with risk and compliance standards.
Guide cross-functional adoption
The role advises teams on where agentic AI does and does not add value, works with business and technology stakeholders, develops reusable practices, and mentors peers.

Likely challenges to think through

Balancing agent autonomy and control
Preparation hypothesis: Agents that take actions across tools may increase automation while making failures harder to contain. Prepare to explain how you would bound actions, make behavior explainable, and evaluate failure modes without assuming a specific Vanguard control process.
Maintaining trust across data sources
Preparation hypothesis: Combining structured and unstructured knowledge from multiple platforms may improve coverage but complicate relevance and trustworthiness. Prepare an example of how you would assess retrieval quality and handle uncertain or conflicting information.
Proving value beyond model scores
Preparation hypothesis: Technical performance may not translate into adoption or useful business outcomes. Prepare to distinguish offline evaluation from measures of coverage, user adoption, and process impact, including when an agent is not the right approach.
Reconciling production and governance needs
Preparation hypothesis: Scalability and performance may compete with security, monitoring, and risk requirements. Prepare to describe how you would work with engineering, risk, and compliance stakeholders to make those trade-offs explicit.

Practice your answers

Technical practice

Suggested practice: When is an AI agent the right solution?

What to coverExplain how you would frame and prioritize a knowledge-intensive business problem before selecting agentic AI, an LLM workflow, or a simpler approach. Address expected business value, the need for autonomous actions, and the costs and risks of added complexity.

Suggested practice: How would you build a trusted enterprise knowledge system?

What to coverDescribe how you would bring structured and unstructured information from multiple business platforms into a system agents can use. Discuss data freshness, provenance, access controls, and what should happen when sources conflict.

Suggested practice: How would you design retrieval for unstructured text?

What to coverWalk through a retrieval-augmented generation approach for a language-driven workflow, including how retrieved material supports an answer. Explain how you would diagnose missed information or answers that are not supported by the source material.

Suggested practice: How should an agent plan and use tools?

What to coverOutline how an agent could complete a multi-step process across tools, workflows, and data sources. Discuss how it should handle tool failures, uncertain results, and actions that require tighter control.

Suggested practice: When would you use multiple agents?

What to coverCompare a multi-agent design with a single tool-augmented agent for a complex process. Explain what separation of responsibilities might achieve and how you would weigh that against coordination, reliability, and evaluation difficulty.

Suggested practice: How would you evaluate business value and model quality?

What to coverPropose an evaluation framework covering coverage, relevance, trustworthiness, explainability, adoption, and technical performance, as the posting requests. Explain how you would connect those measures to a business outcome rather than treating model scores as sufficient.

Suggested practice: How would you monitor an agent after deployment?

What to coverDescribe how you would detect changes in performance, bias, drift, and recurring failure modes. Explain how monitoring findings would lead to investigation, updated evaluations, or changes to the system.

Suggested practice: How would you govern an agent that can act?

What to coverDiscuss guardrails for tool use and generated outputs in an enterprise setting. Explain how you would make decisions and actions reviewable while aligning the design with risk, compliance, and responsible AI standards.

Suggested practice: How would you prepare an LLM solution for production?

What to coverDescribe how you would work with engineers on scalability, security, reliability, and performance. Drawing on the posting's Python and production AI/ML requirements, explain what you would test before release and what operational trade-offs you would track afterward.

Suggested practice: How would you measure whether automation improved a process?

What to coverExplain how you would define a baseline and assess whether an agent changed a meaningful business outcome. Discuss how adoption, process variation, and failures could affect your interpretation of the results.

Behavioural practice

Suggested practice: Tell me about prioritizing an AI opportunity

What to coverUse a real situation in which stakeholders proposed competing problems or solutions. Describe how you framed the options, what you recommended, and the resulting decision or outcome.

Suggested practice: Tell me about resolving a risk or compliance concern

What to coverChoose a real example where a proposed AI or data solution raised a substantive concern. Explain your role, how you worked with the relevant stakeholders, and what changed as a result.

Suggested practice: Tell me about delivering with engineering partners

What to coverDescribe a real cross-functional effort to move an AI/ML solution toward or into production. Cover your actions to resolve a scalability, security, reliability, or performance trade-off and the outcome.

Suggested practice: Tell me about explaining AI limits to a stakeholder

What to coverGive a real example of communicating a technical concept or limitation to a non-technical audience. State the situation, how you adapted your explanation, and what decision or understanding resulted.

Suggested practice: Tell me about mentoring or establishing a reusable practice

What to coverDescribe a real occasion when you helped teammates improve applied AI or business-driven problem solving. Explain what you did, how others used it, and what result you observed.

Sharpen your knowledge

Topics to prepare

Choosing problems for agentic AI
Posting: The role identifies and prioritizes business problems suited to agents, LLMs, and NLP, with measurable outcomes. Suggested exercise: Compare three hypothetical knowledge-intensive workflows; explain which warrants an agent, which needs only retrieval, and which should not use AI.
Enterprise knowledge and RAG design
Posting: Agents must use trusted knowledge drawn from structured and unstructured data across business platforms. Suggested exercise: Sketch how an agent would retrieve and use both data types, then explain how it would handle missing, conflicting, or untrustworthy information.
Agent planning, tool use, and multi-agent trade-offs
Posting: The role designs agents that reason, plan, and act across tools and workflows, including multi-agent solutions. Suggested exercise: Diagram a multi-step workflow and justify when a single tool-augmented agent is preferable to multiple agents.
Business-centric evaluation
Posting: Evaluation must cover coverage, relevance, trustworthiness, explainability, user adoption, technical performance, and business outcomes. Suggested exercise: Draft an evaluation matrix connecting each measure to a failure it would reveal and a decision it would inform.
Production governance and responsible AI
Posting: Deployed systems must address scalability, security, performance, bias, drift, failures, guardrails, risk, and compliance. Suggested exercise: Walk through an agent failure scenario and explain what should be prevented, monitored, and reviewed with engineering and risk stakeholders.

Skills to revise

Python implementation
Posting: The qualifications specify 5+ years of hands-on Python experience. Suggested exercise: Write a small, testable Python workflow that accepts a question, retrieves relevant sample text, and returns an answer with the supporting passage; handle empty or conflicting results.
Unstructured-text processing
Posting: Strong NLP experience with unstructured text and language-driven workflows is required. Suggested exercise: Prepare a small set of differently formatted sample documents, extract useful text, and inspect where formatting or ambiguity degrades retrieval.
Agent workflow prototyping
Posting: The role builds tool-augmented LLM agents that act across data sources and workflows. Suggested exercise: Prototype or pseudocode a bounded, multi-step agent; specify each permitted action, its input and output, and how the workflow stops or handles a failed action.
Evaluation and monitoring practice
Posting: The role implements evaluation, monitoring, and guardrails for LLM and agentic systems. Suggested exercise: Create a small set of normal and failure-case prompts, score answers for relevance and trustworthiness, and define signals that would trigger investigation after deployment.
Production design review
Posting: The role works with engineers to deploy scalable, secure, reliable AI solutions. Suggested exercise: Review a proposed agent design for access to enterprise data, failure handling, performance, and explainability; record the design decisions and unresolved risks.
Stakeholder communication
Posting: The role advises business teams, works across risk, compliance, and technology, and mentors teammates. Suggested exercise: Give two brief explanations of the same proposed use case—one for business stakeholders focused on outcomes and adoption, and one for technical and risk stakeholders focused on evaluation and controls.

Your preparation checklist

Map evidence to the stated qualifications
Prepare a one-page record of relevant work, with dates and specific contributions, covering the posting’s 8 years in data science or related modeling, 3+ years delivering AI/ML in production, 5+ years of hands-on Python, and agent, LLM, and NLP experience. Note any distributed data processing experience separately; the posting calls it a plus.
Select a business-outcome project walkthrough
Prepare one project account that explains the problem, why AI was appropriate, your decisions, cross-functional partners, measurable outcome, and limitations. The role emphasizes business results rather than model performance alone.
Sketch an enterprise knowledge system
As a practice design exercise, draw how an agent could use structured and unstructured data across business platforms. Explain retrieval, source trust, access controls, and how the system handles missing or conflicting information.
Design a bounded agent workflow
Prepare a diagram for a multi-step task showing planning, tool use, orchestration, and any reason to use multiple agents. Mark where actions need validation or human review, and where a simpler non-agent approach would be preferable.
Draft an evaluation plan
Create a sample scorecard covering coverage, relevance, trustworthiness, explainability, user adoption, and technical performance—the measures named in the posting. Include test cases, failure examples, and a proposed way to connect results to a business outcome.
Explain the production and governance approach
Prepare a short deployment walkthrough covering engineering collaboration, scalability, security, performance, monitoring, and guardrails for bias, drift, and failures. Describe how risk and compliance stakeholders would review decisions without assuming a particular Vanguard process.
Prepare leadership examples and employer questions
Choose examples of advising non-technical stakeholders, establishing reusable practices, or mentoring peers, as relevant to your experience. Bring questions about priority agent use cases, available knowledge sources, evaluation ownership, and how business, engineering, risk, and compliance teams collaborate.

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…