New opportunity

Data Scientist

EXL Talent Acquisition Team · Pune, India

About this role

The Data Scientist will develop AI, machine learning and statistical techniques to extract insights and actionable recommendations from large, complex and unstructured datasets. The role includes optimizing and evaluating models, deploying them to production, and collaborating across teams to integrate data-driven solutions.

Responsibilities

Develop and deploy algorithms for complex datasets; design rigorous testing frameworks to validate model performance and improve accuracy and reliability through iteration.
Research and optimize algorithms, methods and technologies to address business problems. Follow advances in data science and share findings with the team.
Establish and maintain data governance policies covering data integrity, security and regulatory compliance.

Qualifications

The job information lists 0–3 years of experience; the qualifications specify 2+ years of prior analytics and data science experience driving AI and advanced analytics projects.
Strong expertise in deep learning frameworks, NLP/text analytics, SVM, LSTM, transformers and neural networks. Hands-on experience with large language models and exposure to fine-tuning open-source models for varied use cases.
Exposure to prompt engineering, vector databases, LangChain and data embeddings. Proficiency in Python for data analysis, machine learning and web development, plus hands-on experience with NumPy, SciPy and scikit-learn.
Strong problem-solving, experimentation and attention to detail, with the ability to communicate effectively with clients and stakeholders.

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 LearningNatural Language ProcessingText AnalyticsSupport Vector MachinesLSTMTransformersNeural NetworksLarge Language ModelsModel Fine-TuningPrompt EngineeringVector DatabasesLangChainData EmbeddingsPythonNumPySciPyscikit-learnStatistical AlgorithmsModel EvaluationData GovernanceStakeholder Communication

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

Turn complex data into usable solutions
The role calls for AI and statistical algorithms that extract actionable recommendations from large, complex, unstructured data, then move into production. Prepare to explain how an analysis becomes a deployable solution tied to a business problem.
Validate and improve model performance
Rigorous testing, result validation, and iteration to improve accuracy and reliability are explicit duties. Prepare a concrete account of evaluation choices, what results showed, and how those results guided changes.
Apply NLP and large language models
The qualifications emphasize deep learning, NLP/text analytics, transformers, and hands-on LLM work, with exposure to open-source fine-tuning, prompts, embeddings, vector databases, and LangChain. Prepare to explain when different approaches are appropriate rather than listing tools.
Maintain data governance
The role includes establishing and maintaining policies for data integrity, security, and regulatory compliance. Prepare to discuss how those concerns affect data and model work without assuming a particular regulatory framework.
Research and collaborate across teams
The posting calls for exploring methods, sharing findings, integrating data-driven solutions across teams, and communicating with diverse clients and stakeholders. Prepare to explain how technical findings inform a practical recommendation.

Likely challenges to think through

Balancing experimentation with dependable results
A likely difficulty is deciding when a promising method has enough evidence to justify further development or deployment. Prepare to compare candidate approaches using a clear test design, reliability criteria, and an explanation of the trade-offs.
Making advanced methods useful for a business problem
The breadth of listed NLP and LLM techniques may make method selection difficult: greater complexity does not necessarily produce a more useful recommendation. Prepare to explain how you would select an approach based on the data, intended use, and evaluation results.
Integrating solutions while preserving governance
Production deployment and cross-team integration may need to be reconciled with data integrity, security, and compliance duties. Prepare to describe how you would identify governance constraints early and communicate their effect on a proposed solution.

Practice your answers

Technical practice

How would you turn a large, unstructured dataset into actionable recommendations?

What to coverSuggested preparation: Explain how you would define the business question, assess data quality, prepare the data and select an analytical approach. Describe how you would check that the resulting insights support a useful recommendation rather than merely a predictive score.

When would you choose an SVM, LSTM or transformer for a text analytics problem?

What to coverSuggested preparation: Compare their suitability for the task and available data, including trade-offs in complexity and computational cost. Explain how you would establish a baseline and test whether a more complex model improves results.

How would you design a rigorous framework for evaluating a model?

What to coverSuggested preparation: Discuss appropriate data splits, metrics tied to the business objective, and checks for leakage or performance differences across relevant data slices. Explain how you would use the results to guide iteration on accuracy and reliability.

How would you decide whether to fine-tune an open-source large language model?

What to coverSuggested preparation: Compare fine-tuning with prompting for a defined use case, considering available examples, evaluation results and implementation effort. Explain how you would test whether fine-tuning improves the behavior that matters for that use case.

How would you improve an LLM's behavior through prompt engineering?

What to coverSuggested preparation: Describe a controlled way to vary instructions and examples while keeping evaluation cases consistent. Address how you would inspect failures and avoid treating a few convincing responses as evidence of reliable performance.

How do embeddings, a vector database and LangChain fit into a text-retrieval solution?

What to coverSuggested preparation: Explain the distinct roles of text embeddings, similarity search and application orchestration without assuming a particular architecture is required. Discuss how you would evaluate whether retrieved material is relevant and improves the final answer.

What would you check before deploying a model to production?

What to coverSuggested preparation: Cover validation against the intended use, reproducible inputs and outputs, and how you would detect a decline in performance after deployment. Explain the trade-off between improving a model further and releasing a sufficiently tested version.

How would you use Python, NumPy, SciPy and scikit-learn in an analysis workflow?

What to coverSuggested preparation: Give concrete examples of array-based preparation, statistical analysis and model fitting or evaluation that fit each library's purpose. Explain how you would make the work reproducible and verify intermediate results.

How would you protect data integrity, security and compliance in a data science project?

What to coverSuggested preparation: Discuss checks on data provenance and quality, appropriate access controls, and documenting how data is used. Avoid assuming a particular regulation; explain how you would establish which obligations apply before setting governance rules.

How would you assess a new method for improving an existing AI model?

What to coverSuggested preparation: Define the observed limitation, a measurable hypothesis and a comparison against the current approach. Discuss how you would weigh gains in performance against complexity and share useful findings with the team.

Behavioural practice

Tell me about a time you explained a technical finding to a client or stakeholder.

What to coverSuggested preparation: Use a real situation, the audience's decision, and the actions you took to make the finding understandable. State the result, including how you handled uncertainty or questions.

Describe a time an experiment or model underperformed and you iterated on it.

What to coverSuggested preparation: Give the initial goal and evidence that the approach fell short, then explain how you diagnosed the issue and chose the next test. Report the outcome, including what you learned if performance did not improve.

Tell me about a time you addressed a data-quality, security or governance concern.

What to coverSuggested preparation: Describe the concern, your responsibility and the steps you took with relevant colleagues. Explain the result without implying that every issue could be resolved by model changes alone.

Describe a time you worked across teams to integrate a data-driven solution.

What to coverSuggested preparation: Set out the teams' differing needs, what you did to coordinate the work and how decisions were made. Explain the outcome and your specific contribution.

Tell me about a time you shared a useful data science finding with your team.

What to coverSuggested preparation: Identify the method or insight, why it was relevant and how you evaluated it before sharing it. Describe what the team did with the information, if anything, and what you learned.

Sharpen your knowledge

Topics to prepare

Model evaluation and iteration
Revision exercise: explain how you would split a complex dataset, choose performance measures, validate results and investigate an apparent improvement. The posting explicitly calls for rigorous testing and iteration to improve accuracy and reliability.
Statistical and machine-learning methods
Revision exercise: compare when an SVM, a statistical algorithm and a neural network might suit the same business problem, including their limitations. The posting names SVMs, statistical algorithms and algorithm optimization.
Deep learning for text
Revision exercise: contrast LSTMs and transformers for an NLP or text-analytics task, including how you would assess their outputs. The posting explicitly seeks expertise in deep learning, NLP, LSTMs and transformers.
LLMs and retrieval concepts
Revision exercise: explain the distinct roles of prompting, open-source model fine-tuning, embeddings, a vector database and LangChain in a text use case. All are named in the qualifications; combining them into one design is a preparation suggestion, not a stated project requirement.
Data governance
Revision exercise: describe how you would check data integrity and address security and regulatory compliance before using a complex dataset. Establishing and maintaining governance policies is an explicit responsibility.

Skills to revise

Build a reproducible Python baseline
Practice suggestion: use Python with NumPy, SciPy and scikit-learn to prepare sample data, fit a baseline and report its evaluation results. These libraries and Python for analysis and machine learning are explicitly listed.
Run a controlled model comparison
Practice suggestion: hold a validation approach constant while comparing two model choices or settings; record the result and a plausible reason for any difference. This rehearses the posting’s testing, experimentation and optimization responsibilities.
Inspect text-model behavior
Practice suggestion: take a small text example and document where an NLP model succeeds or fails, then propose one targeted change to test. This is relevant to the stated text-analytics expertise and emphasis on iterating to improve reliability.
Explain a production handoff
Practice suggestion: sketch how a trained model could be made available through Python web code, then identify what you would validate before deployment. The posting mentions Python web development and deploying models to production; it does not specify a serving framework.
Communicate an actionable recommendation
Practice suggestion: summarize a model result for a nontechnical stakeholder, stating the recommended action, supporting evidence and an important limitation. The role calls for actionable recommendations and effective communication with clients and stakeholders.

Your preparation checklist

Prepare an AI/analytics project walkthrough
The posting calls for experience driving AI and advanced analytics projects. Prepare one concise account of the problem, complex or unstructured data, your contribution, model choices, results and actionable recommendations.
Document a model-evaluation example
The role emphasizes rigorous testing and iteration. Bring a comparison of a baseline and an improved model, including validation design, relevant metrics, error analysis and how you checked reliability.
Practice an NLP and LLM design exercise
The qualifications name NLP, transformers, LLMs, open-source model fine-tuning, prompts, embeddings, vector databases and LangChain. Sketch a solution to a text-data problem, explaining which components you would use, alternatives and how you would evaluate outputs.
Prepare evidence of hands-on tools and methods
Make a short reference sheet pointing to examples of Python, NumPy, SciPy, scikit-learn and relevant deep-learning work. Be ready to explain any SVM, LSTM or neural-network choices you have actually made; Python web development is also listed.
Map a model’s path to production
The role includes deployment and cross-team integration. As a preparation exercise, draw a simple workflow from data intake through testing, deployment and monitoring, noting assumptions and decisions that would need input from other teams.
Draft governance and stakeholder questions
The posting assigns data integrity, security and regulatory-compliance responsibilities and calls for client communication. Prepare an example of how you explained a technical finding, plus questions about data controls, stakeholders and how the team validates deployed models.
Clarify the experience expectations
The page lists 0–3 years under job information but specifies 2+ years in the qualifications. Prepare a question about how the team interprets those two statements for this Data Scientist role.

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…