New opportunity

Affle India - Data Scientist - LLM/RAG

AFFLE 3I LIMITED · Gurgaon, India

About this role

The Data Scientist will lead data science initiatives that address business problems and improve advertising campaign performance. Based in the Gurgaon office, the role combines advanced analytics, machine learning, LLM/RAG systems and collaboration with business stakeholders.

Responsibilities

Translate stakeholder goals into data science projects; lead and mentor the team and communicate actionable findings to non-technical stakeholders and senior leadership.
Develop scalable, robust ML algorithms for ad placement within privacy frameworks. Optimize budget allocation under high cardinality and uncertainty, and build and optimize recommendation systems for campaign performance.
Explore complex datasets, conduct statistical analyses and develop ML/deep learning models for forecasting and classification, balancing performance with interpretability. Turn predictive-modeling proofs of concept into production-ready solutions.
Design, deploy and maintain AI agents and RAG-based systems. Build orchestration workflows with LangChain, LangGraph or similar tools, and use advanced SQL for analysis, feature engineering and reporting.

Requirements

A bachelor's or master's degree in a quantitative field such as Computer Science, Statistics, Physics or Mathematics, and 3+ years of relevant data science and machine learning experience with a focus on leading projects. The listing also displays a 3–6-year experience range.
Strong Python coding for production-grade development and deployment; expert Python and SQL proficiency, including query and data-processing optimization. A solid background in statistical analysis, hypothesis testing, experimental design and causal inference is required.
Exposure to RAG, LLMs and agentic workflows, including fine-tuning, evaluation and prompt engineering; practical familiarity with LangChain, LangGraph or similar orchestration toolkits. Strong problem-solving, communication and presentation skills are required.

Preferred

Experience with Google Cloud Platform services, including BigQuery, AI Platform and Vertex AI. A background in AdTech, particularly optimization and bidding strategies, is a plus.

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

Skills for this role

Data ScienceMachine LearningDeep LearningPythonSQLExploratory Data AnalysisStatistical AnalysisHypothesis TestingExperimental DesignCausal InferenceForecastingClassificationPredictive ModelingRecommendation SystemsFeature EngineeringRAGLarge Language ModelsAI AgentsFine-tuningModel EvaluationPrompt EngineeringLangChainLangGraphGoogle Cloud PlatformBigQueryAI PlatformVertex AICommunicationMentoring

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

Define and lead business-focused data science work
Translate stakeholder goals into projects, guide and mentor data scientists, and communicate actionable findings to non-technical stakeholders and senior leadership.
Improve ad placement and budget allocation
Lead scalable, robust ad-placement models within privacy frameworks and optimize budgets where choices are high-cardinality and outcomes uncertain.
Build models that improve campaign performance
Develop forecasting, classification and recommendation systems, using exploratory analysis and statistical methods while balancing predictive performance with interpretability.
Deliver RAG and agentic systems
Design, deploy and maintain AI agents and RAG systems, including orchestration workflows using LangChain, LangGraph or similar tools.
Move analysis and prototypes into production
Turn predictive-modeling proofs of concept into production-ready solutions, supported by production-grade Python and optimized SQL for analysis, feature engineering and reporting.

Likely challenges to think through

Optimizing under privacy and uncertainty
Likely preparation challenge: the stated ad-placement and budget-allocation duties combine privacy constraints, many possible choices and uncertain outcomes. Prepare to explain how you would frame the objective, evaluate trade-offs and check that an approach remains scalable; this is not a claim about Affle's current systems.
Showing impact without sacrificing interpretability
Likely preparation challenge: a more predictive model may be harder to explain, while observed campaign gains may not establish causation. Prepare an example of choosing evaluation measures and using experimental design or causal reasoning to support an actionable recommendation.
Taking RAG or agent workflows beyond a demo
Likely preparation challenge: a workflow that works as a proof of concept may be harder to evaluate, deploy and maintain. Prepare to describe how you would test response quality and failures, choose orchestration boundaries, and plan for production maintenance; the posting does not specify a particular implementation.

Practice your answers

Technical practice

How would you design an ad-placement model that remains effective within privacy constraints?

What to coverSuggested preparation: explain how you would define the placement objective, identify permissible inputs, and evaluate model quality alongside scalability and robustness. Discuss how privacy constraints could change feature choices and measurement.

How would you allocate campaign budgets when there are many possible choices and outcomes are uncertain?

What to coverSuggested preparation: describe an optimization objective, constraints, and a way to estimate uncertain returns. Explain how you would compare an allocation strategy with a baseline and respond when estimates change.

How would you explore a complex dataset before recommending a data science initiative?

What to coverSuggested preparation: cover data quality, distributions, missingness, segmentation, and patterns that warrant further testing. Distinguish an observed trend from evidence strong enough to support a business recommendation.

How would you choose and evaluate a forecasting or classification model?

What to coverSuggested preparation: connect the model choice and validation approach to the business decision and available data. Discuss performance versus interpretability, including how you would detect leakage or unreliable results.

How would you test whether a campaign change caused an improvement?

What to coverSuggested preparation: explain hypothesis definition, experimental design, outcome metrics, and statistical uncertainty. If an experiment is impractical, describe a causal-inference approach and the assumptions that would limit its conclusions.

How would you build and improve a recommendation system for campaign performance?

What to coverSuggested preparation: define what is being recommended, how success would be measured, and how you would establish a baseline. Discuss trade-offs among relevance, coverage, changing data, and the cost of serving recommendations.

How would you use and optimize SQL for analytical features and reporting?

What to coverSuggested preparation: explain how you would structure a complex query, check its correctness, and avoid expensive or unnecessary processing. Discuss the trade-off between reusable feature definitions and query performance; BigQuery is a preferred platform in the posting, not a required one.

How would you evaluate a RAG system and decide whether prompting or fine-tuning is warranted?

What to coverSuggested preparation: separate retrieval quality from answer quality and describe how you would assess groundedness, relevance, and failure cases. Explain what evidence would justify changing prompts, retrieval, or the model rather than assuming fine-tuning is needed.

How would you orchestrate and maintain an AI-agent workflow?

What to coverSuggested preparation: outline the responsibilities of workflow steps and how a toolkit such as LangChain or LangGraph could coordinate them. Discuss testing, failure handling, observability, and the trade-off between flexible agent behavior and predictable operation.

How would you turn a predictive-modeling proof of concept into a production-ready solution?

What to coverSuggested preparation: describe the checks needed before deployment, including reliable Python code, reproducible data processing, and validation against the original business goal. Explain how you would monitor quality and maintain the solution; the posting lists GCP services as preferred experience, not a required deployment stack.

Behavioural practice

Tell me about a time you translated an unclear stakeholder goal into a data science project.

What to coverSuggested preparation: use a real situation, your actions to clarify the decision and scope, and the resulting project direction. Explain how you handled competing interpretations of success.

Describe a time you led a technical project while mentoring a less experienced colleague.

What to coverSuggested preparation: give the situation, the responsibilities you took, and specific guidance or feedback you provided. State the result for both the work and the colleague without overstating your contribution.

Tell me about a time you presented a complex analytical finding to non-technical leadership.

What to coverSuggested preparation: describe the decision at stake, how you explained uncertainty and made the finding actionable, and what happened next. Focus on your communication choices rather than technical detail alone.

Describe a time you made progress on an ambiguous data problem.

What to coverSuggested preparation: identify what was unknown, the steps you took to test assumptions, and the result. Explain which uncertainties remained and how you communicated them.

Tell me about a time you helped move a data science approach from exploration toward wider use.

What to coverSuggested preparation: use a real situation to explain how you assessed readiness, worked with others, and addressed objections or risks. State the outcome, including any limits on adoption or production use.

Sharpen your knowledge

Topics to prepare

Ad placement, budget allocation and recommendations
Posting: Lead scalable ad-placement models within privacy frameworks, optimize budgets under high cardinality and uncertainty, and improve campaign recommendations. Suggested exercise: Sketch a hypothetical campaign decision problem, its constraints and evaluation criteria; explain how uncertainty and privacy constraints would change the approach. Bidding-strategy experience is a plus, not a requirement.
Experimental design and causal inference
Posting: Advanced statistical analysis, hypothesis testing, experimental design and causal inference are required. Suggested exercise: Design a hypothetical campaign experiment, state the hypothesis and outcome measure, then explain what would prevent a causal conclusion.
Forecasting and classification trade-offs
Posting: Lead ML/DL modeling for forecasting and classification while balancing performance with interpretability. Suggested exercise: Compare two plausible model approaches for a hypothetical campaign outcome and explain how you would assess both predictive quality and interpretability.
RAG and agent-system design
Posting: Design, deploy and maintain RAG systems and AI agents; exposure to fine-tuning, evaluation and prompt engineering is requested. Suggested exercise: Diagram a hypothetical RAG-and-agent workflow and explain where you would evaluate retrieval, generated answers and agent behavior.
Project definition and production readiness
Posting: Translate stakeholder goals into projects, lead work, and move predictive-modeling proofs of concept toward production-ready solutions. Suggested exercise: Turn an ambiguous campaign-performance goal into a scoped project, then identify what must be checked before a proof of concept is deployed.

Skills to revise

Production-grade Python
Posting: Strong Python coding for production-grade development and deployment is required. Suggested exercise: Refactor a small predictive-modeling script into reusable, testable code and explain how you would handle invalid inputs and deployment changes.
Advanced SQL and query optimization
Posting: Expert SQL is requested for analysis, feature engineering and reporting, including query optimization. Suggested exercise: Write a query that builds campaign-level features from hypothetical event data, then explain how you would check correctness and improve its efficiency.
Exploratory analysis and model evaluation
Posting: Conduct in-depth data exploration and develop forecasting or classification models. Suggested exercise: On a practice dataset, document data-quality issues and patterns, build a baseline model, and explain which results support or limit a recommendation.
LangChain or LangGraph orchestration
Posting: Practical experience with LangChain, LangGraph or a similar toolkit is requested for scalable AI workflows. Suggested exercise: Build a small RAG workflow with one of those toolkits, then test how it responds to a missing or irrelevant retrieved passage.
Stakeholder communication and mentoring
Posting: Communicate actionable findings to non-technical stakeholders and senior leadership, and mentor junior data scientists. Suggested exercise: Give a brief, plain-language presentation of a hypothetical modeling result, including its uncertainty; then explain how you would review a junior colleague's analysis.
Preferred GCP services
Posting: Experience with GCP services including BigQuery, AI Platform and Vertex AI is preferred, not required. Optional exercise: Review how a BigQuery analysis could support a modeling workflow and outline where AI Platform or Vertex AI might fit, without assuming this employer uses a particular deployment process.

Your preparation checklist

Map the role to preparation examples
The posting emphasizes leading data science projects and mentoring. Prepare a one-page map of relevant projects you can discuss, noting your decisions, collaborators, outcomes and any mentoring; leave out experience you cannot substantiate.
Prepare a project walkthrough
Choose one relevant project, if available, and outline the business goal, data exploration, model choices, evaluation, deployment decisions and results. Practice explaining what you personally led and what you would change.
Work through an ad-campaign design exercise
Sketch a hypothetical approach to privacy-aware ad placement, budget allocation under uncertainty, or campaign recommendations. State assumptions, objectives, constraints, evaluation measures and scalability trade-offs; these are preparation exercises, not stated company projects.
Rehearse statistical decision-making
Prepare an example or hypothetical plan covering a hypothesis test, experimental design and causal-inference risks. Explain how you would distinguish a campaign improvement from confounding and communicate uncertainty.
Demonstrate Python and SQL depth
Prepare a small code or query example showing feature engineering, query or processing optimization, tests and production-readiness considerations. The posting specifically calls for production-grade Python and advanced SQL.
Draw a RAG and agent architecture
Create a design sketch for retrieval, prompts, orchestration with LangChain, LangGraph or a similar toolkit, evaluation, deployment and maintenance. Be ready to explain when fine-tuning would or would not help; do not imply the posting specifies a particular use case.
Practice an executive-level explanation
Turn one technical finding into a brief presentation with the business decision, evidence, limitations and recommended next step. Include how you would define the project with stakeholders and guide junior scientists.
Bring targeted questions
Ask how this team prioritizes campaign-performance problems, evaluates RAG and ML systems, handles privacy constraints, and moves proofs of concept into production. Ask which GCP services are used in practice; GCP experience is listed as preferred.

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…