New opportunity

Senior Data Scientist

Kroll · Manila, Philippines

About this role

Kroll is hiring a Senior Data Scientist for its Enterprise Data Group. The role leads end-to-end machine learning initiatives, mentors junior colleagues, and works with business and engineering stakeholders to turn complex problems into production-grade solutions. The team's work spans fintech products, digital transformation, ML-based process automation, business intelligence, data governance, and generative AI.

Responsibilities

Design, research, implement, and evaluate traditional machine learning, deep learning, NLP, and LLM/GenAI solutions. Build and fine-tune models ranging from statistical models and gradient-boosted trees to transformers and retrieval-augmented generation systems.
Develop prompts, evaluation frameworks, and guardrails for LLM applications. Build scalable data and ML pipelines in Databricks using PySpark, Delta Lake, and MLflow.
Deploy and maintain production models on Azure, including CI/CD, versioning, monitoring, and drift detection. Validate inputs, outputs, and business impact, and establish testing practices.
Scope problems with engineering, product, and business partners; communicate results and tradeoffs to clients and senior leaders; mentor junior data scientists and contribute to standards for code quality, experimentation, and responsible AI.

Requirements

An MS or PhD in computer science, statistics, mathematics, analytics, or a related quantitative field, plus at least 5 years of applied machine learning experience, including production model delivery.
Strong Python skills; experience with scikit-learn, PyTorch or TensorFlow, pandas, and Hugging Face Transformers; hands-on Databricks and Spark/PySpark experience; and production experience on Azure, ideally with Azure AI Foundry, Azure OpenAI Service, and Azure Data Lake.
Breadth across statistical ML, deep learning, NLP, and LLM/GenAI, including practical work with prompts, RAG, fine-tuning, embeddings, vector databases, evaluation, and agentic workflows. Knowledge of the full ML lifecycle, structured and unstructured data, statistics, uncertainty, bias, and model risk, with strong technical and business communication skills.

Preferred

Financial services, risk, compliance, or regulatory experience; MLOps tools such as Docker, Kubernetes, Azure DevOps, or GitHub Actions; agentic frameworks such as LangChain, LlamaIndex, or Semantic Kernel; LLM evaluation and production GenAI deployment; and responsible AI practices covering fairness, explainability, and privacy. The posting identifies the arrangement as hybrid and lists Manila, Philippines, plus five other locations that are not named.

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

Skills for this role

PythonMachine LearningDeep LearningNatural Language ProcessingGenerative AILarge Language ModelsRetrieval-Augmented GenerationPrompt EngineeringEmbeddingsAgentic AIDatabricksPySparkApache SparkDelta LakeMLflowUnity CatalogMicrosoft AzureAzure AI FoundryAzure OpenAI ServiceAzure FunctionsAzure Kubernetes ServiceAzure Data LakeCI/CDModel MonitoringDrift DetectionData ValidationFeature EngineeringModel EvaluationStatistical ModelingGradient-Boosted Treesing in our Data Science team. You will work alongside anq4

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

Lead ML work end to end
The role owns research, model design, experimentation, evaluation, deployment, and monitoring across traditional ML, deep learning, NLP, and GenAI.
Make LLM applications testable
The posting calls for prompts, RAG and fine-tuning, alongside evaluation frameworks and guardrails for LLM-powered applications.
Build production data and model pipelines
Responsibilities include scalable Databricks pipelines using PySpark, Delta Lake, and MLflow, plus Azure deployment with CI/CD, versioning, monitoring, and drift detection.
Connect ML work to measurable outcomes
The scientist scopes problems with engineering, product, and business partners; validates business impact; and explains results and trade-offs to clients and senior leaders.
Raise team standards
The role mentors junior data scientists and contributes to standards for code quality, experimentation, and responsible AI.

Likely challenges to think through

Choosing proportionate solutions
Preparation hypothesis: The range from statistical models to RAG creates a trade-off between capability, complexity, and maintainability. Prepare an example of how you would compare approaches against a baseline and select one using evaluation results.
Keeping models dependable after launch
Preparation hypothesis: Moving from experimentation to Azure production may make data validation, reproducibility, and drift harder to manage. Prepare a clear account of testing, versioning, monitoring, and response to degraded performance.
Balancing impact with model risk
Preparation hypothesis: Work connected to governance and risk may require careful treatment of uncertainty, bias, privacy, and explainability while still delivering useful outcomes. Prepare to explain those trade-offs in business terms without assuming a specific client or regulatory requirement.

Practice your answers

Technical practice

Suggested practice: How would you scope an ML problem with product, engineering, and business partners?

What to coverPreparation suggestion: Explain how you would define the decision the model supports, establish a baseline, and agree on measurable business outcomes. Discuss when a simpler statistical model may be preferable to a more complex approach.

Suggested practice: How would you build a reliable pipeline for structured data and documents in Databricks?

What to coverPreparation suggestion: Cover input validation, PySpark transformations, Delta Lake data management, and reproducible jobs. Explain how you would handle changing schemas or poor-quality source data.

Suggested practice: How would you evaluate a gradient-boosted model before production?

What to coverPreparation suggestion: Discuss baselines, leakage prevention, appropriate validation splits, and metrics tied to the business problem. Address uncertainty, bias, and the trade-off between predictive performance and interpretability.

Suggested practice: When would you fine-tune a transformer for an NLP task?

What to coverPreparation suggestion: Compare fine-tuning with a simpler model or prompting, considering labeled data, cost, and expected improvement. Describe evaluation on representative text and error analysis before deployment.

Suggested practice: How would you design and assess a RAG system for document-based questions?

What to coverPreparation suggestion: Explain document preparation, embeddings, retrieval through a vector database, and how retrieved context reaches the model. Discuss tests for retrieval quality, answer grounding, and failure cases.

Suggested practice: How would you evaluate prompts and guardrails for an LLM application?

What to coverPreparation suggestion: Describe a repeatable test set and measures for answer quality, unsafe outputs, and unsupported claims. Explain how prompt changes and guardrails would be compared without relying only on appealing examples.

Suggested practice: When would an agentic workflow be justified over a simpler LLM workflow?

What to coverPreparation suggestion: Compare the value of multi-step tool use with added complexity, latency, and failure modes. Explain how you would constrain actions, test behavior, and detect errors; the posting identifies agentic workflows and lists specific frameworks as preferred experience.

Suggested practice: How would you keep Databricks ML experiments reproducible?

What to coverPreparation suggestion: Discuss tracking data and model versions, parameters, metrics, and artifacts with MLflow. Explain how experiment results would inform promotion of a model rather than treating a single metric as decisive.

Suggested practice: How would you deploy and maintain a model on Azure?

What to coverPreparation suggestion: Cover versioning, CI/CD checks, deployment, and monitoring of inputs, outputs, and drift. Explain what evidence would trigger investigation, rollback, or retraining.

Suggested practice: How would you assess model risk and business impact after release?

What to coverPreparation suggestion: Distinguish model-quality measures from the outcome the solution is meant to improve. Address testing and monitoring for fairness, explainability, privacy, and unintended effects where relevant to the use case.

Behavioural practice

Suggested practice: Tell me about scoping an ambiguous problem with stakeholders.

What to coverPreparation suggestion: Use a real situation to explain differing goals, your actions to define a workable ML objective, and the result. Clarify how you reached agreement with business and engineering partners.

Suggested practice: Tell me about explaining a technical trade-off to a non-technical audience.

What to coverPreparation suggestion: Describe the audience and decision, how you presented uncertainty and alternatives, and the resulting decision or feedback. A client or senior-leadership example is especially relevant if you have one.

Suggested practice: Tell me about responding to a production-model issue.

What to coverPreparation suggestion: Give a real situation involving a quality, data, or monitoring concern; describe your diagnosis and coordination with others. State the result and what changed in testing or monitoring afterward.

Suggested practice: Tell me about mentoring a junior data scientist.

What to coverPreparation suggestion: Explain the colleague's development need, the guidance or review practices you used, and the observable result. Focus on their growth and any improvement to shared code or experimentation standards.

Suggested practice: Tell me about handling a responsible-AI concern in an ML project.

What to coverPreparation suggestion: Describe a real concern involving bias, explainability, privacy, or model risk, the actions you took with relevant partners, and the outcome. Be clear about any unresolved trade-offs rather than claiming the risk was eliminated.

Sharpen your knowledge

Topics to prepare

Model choice and statistical judgment
The posting spans classical statistical models, gradient-boosted trees, deep learning, and transformers, and calls for reasoning about uncertainty and bias. Suggested exercise: compare when each approach would suit a structured-data or document problem, and explain the evaluation metrics, uncertainty, and tradeoffs you would report.
LLM application design and evaluation
The role explicitly covers prompts, RAG, fine-tuning, embeddings, vector databases, agentic workflows, evaluation, and guardrails. Suggested exercise: diagram a RAG application, then explain when retrieval or fine-tuning would help and how you would test answer quality and guardrail failures.
Production ML lifecycle
Kroll asks for data validation, feature engineering, experimentation, deployment, versioning, monitoring, and drift detection. Suggested exercise: trace a model from incoming data to production and name a test, failure mode, and response at each stage.
Responsible AI and model risk
The requirements mention bias and model risk; fairness, explainability, and privacy are preferred knowledge. Suggested exercise: review a hypothetical model decision for possible bias, sensitive-data exposure, and explainability limits, distinguishing measurable checks from unresolved risks.
Business scoping and technical leadership
The role involves measurable outcomes, communication with clients and senior leaders, mentoring, and team standards. Suggested exercise: turn a hypothetical stakeholder request into a problem statement, success measure, experiment plan, and brief explanation of tradeoffs for a non-technical audience.

Skills to revise

Python model development
Strong Python and experience with pandas, scikit-learn, PyTorch or TensorFlow, and Hugging Face Transformers are specified. Suggested exercise: use a small sample dataset to prepare features, establish a scikit-learn baseline, evaluate a second model, and explain what would justify a more complex approach.
Databricks data and experiment pipelines
The posting calls for Databricks notebooks and jobs, Spark/PySpark, Delta Lake, MLflow, and Unity Catalog. Suggested exercise: sketch or build a PySpark pipeline that validates input, writes a Delta table, records an MLflow experiment, and specifies how a Databricks job and Unity Catalog would fit.
RAG prototyping and testing
Hands-on prompt engineering, RAG, embeddings, vector databases, and LLM evaluation are required. Suggested exercise: prototype retrieval over a small document set, write prompts that use retrieved passages, and test answers for relevance, unsupported claims, and guardrail failures.
Azure deployment and monitoring plan
Production Azure experience is required; the responsibilities name Azure AI Foundry, Azure OpenAI, Azure Functions, and AKS, alongside CI/CD, versioning, and drift detection. Suggested exercise: draft a deployment and rollback plan for a model endpoint, including input/output tests, version tracking, monitoring signals, and a drift response.
Explaining results and coaching peers
The role requires communicating tradeoffs to non-technical audiences and mentoring junior data scientists. Suggested exercise: prepare a two-minute explanation of an experiment’s business impact and limitations, then practice giving specific review feedback on its tests, reproducibility, and code quality.

Your preparation checklist

Map your evidence to the requirements
Prepare a one-page inventory of examples you can substantiate for the advanced degree, 5+ years of applied ML, production delivery, Python, Databricks/PySpark, and Azure. Mark gaps rather than inventing examples.
Prepare an end-to-end model walkthrough
Choose a real project and outline problem scoping, data validation, feature engineering, model selection, testing, deployment, monitoring, and measured business impact. The posting asks for production-grade ML delivery.
Rehearse model and statistical tradeoffs
Prepare a comparison from work you have done, such as a statistical model or gradient-boosted trees versus a deep-learning approach. Explain evaluation choices, uncertainty, bias, and model risk.
Sketch a GenAI design exercise
Practice designing a document-based RAG application, since the role covers RAG and unstructured data. Show how you would choose retrieval and prompts, evaluate outputs, add guardrails, and decide whether fine-tuning is warranted; label this as a proposed design, not past work.
Diagram a production pipeline
Draw how you would use Databricks, PySpark, Delta Lake, and MLflow with Azure deployment. Be ready to discuss CI/CD, versioning, input and output checks, monitoring, and drift detection—the posting names these responsibilities.
Prepare leadership examples and employer questions
Outline a real example of scoping with business and engineering partners, explaining tradeoffs to non-technical stakeholders, or mentoring a colleague. Bring questions about the Enterprise Data Group’s priorities, production ownership, evaluation standards, and how it measures impact.

Availability last checked: Nov 5, 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 Philippines. Matched to this role, not your profile.

Explore more jobs

Finding similar opportunities…