New opportunity

Data Scientist

SAP · Singapore

About this role

SAP is seeking a Data Scientist in Singapore to deliver end-to-end AI solutions for customer support. The role works across machine learning, LLM-based applications, retrieval and recommendation systems, and data-driven services integrated into support processes. The working arrangement is hybrid, with expected travel of 0–10%.

Responsibilities

Understand business and technical requirements in the Support Delivery domain and translate them into scalable AI solutions with technical and business stakeholders.
Build AI-powered services for case analysis, knowledge retrieval, recommendations, classification, and automation.
Move models and algorithms from experimentation into production with machine learning engineers and developers. Design data pipelines, model pipelines, and ML microservices where needed.
Analyze large-scale structured and unstructured data to identify patterns, improve model performance, and generate actionable insights. Apply LLMs, embeddings, RAG, semantic search, knowledge graphs, and agentic workflows where relevant.
Collaborate with data scientists, data engineers, software developers, and product stakeholders to deliver production-ready solutions.

Requirements

3+ years of experience in machine learning or applied AI as a hands-on Data Scientist, plus a relevant university degree in Data Science, AI, or an equivalent field.
Strong Python and practical data science skills, including data analysis, feature engineering, model development, evaluation, and experimentation with structured or unstructured data.
Strong problem-solving and communication skills, curiosity, independence, and the ability to collaborate with engineering teams.

Valuable familiarity includes SQL, cloud-native tools, production readiness, Spark, or Elasticsearch. The team builds AI-native microservices, applications, and agentic workflows for customer support. SAP notes learning and growth opportunities, benefits, and a focus on well-being; successful candidates might be required to undergo external background verification.

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

Skills for this role

PythonMachine LearningApplied AIData AnalysisFeature EngineeringModel DevelopmentModel EvaluationExperimentationSQLSparkElasticsearchLarge Language ModelsEmbeddingsRAGSemantic SearchKnowledge GraphsAgentic WorkflowsRecommendation SystemsData PipelinesModel PipelinesML MicroservicesStructured Data AnalysisUnstructured Data AnalysisProblem-SolvingCommunicationCollaboration

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 support needs into AI solutions
The role calls for understanding business and technical requirements in Support Delivery and working with both groups to turn them into scalable solutions.
Build customer-support services
The advertised use cases include case analysis, knowledge retrieval, recommendations, classification, and automation integrated into support processes.
Move experiments into production
SAP expects collaboration with machine learning engineers and developers to turn models into reliable solutions, including data pipelines, model pipelines, and ML microservices where needed.
Use data to improve results
The work includes analyzing large-scale structured and unstructured data, evaluating models, identifying patterns, and generating actionable insights.
Choose AI techniques for the use case
The posting names LLMs, embeddings, RAG, semantic search, knowledge graphs, and agentic workflows, with the qualification that they are applied where relevant.

Likely challenges to think through

Balancing retrieval quality and solution complexity
Preparation hypothesis, not a stated SAP problem: support knowledge retrieval may benefit from advanced techniques without needing every technique listed. Prepare to explain how you would compare simpler retrieval with RAG or semantic search against the support task and evaluation results.
Evaluating mixed support data
Preparation hypothesis: structured and unstructured data can make consistent model evaluation difficult. Prepare an example of how you defined a useful outcome, checked data quality, and used experiments to decide whether a model improved it.
Preserving value beyond the prototype
Preparation hypothesis: an experimental model may need changes before it can run reliably within a support process. Prepare to discuss trade-offs between model improvement and production readiness, including how you would coordinate interfaces and handoffs with engineers.

Practice your answers

Technical practice

How would you turn a customer-support problem into an AI solution?

What to coverThe posting calls for translating business and technical requirements in Support Delivery into scalable solutions. Suggested preparation: explain how you would define the task, establish a baseline, agree on success measures with stakeholders, and decide whether AI is warranted.

How would you prepare structured and unstructured support data for modeling?

What to coverThe role includes large-scale data analysis and feature engineering in Python. Suggested preparation: discuss data quality checks, useful features, leakage risks, and how your approach would differ for records and free text.

How would you build and evaluate a support-case classifier?

What to coverClassification, model development, and evaluation are stated parts of the role. Suggested preparation: compare a simple baseline with a more complex model, justify metrics in light of different error costs, and explain how you would inspect failures.

How would you design knowledge retrieval using embeddings and RAG?

What to coverThe posting names knowledge retrieval, embeddings, semantic search, and RAG. Suggested preparation: explain document preparation, retrieval and ranking choices, how retrieved material reaches an LLM, and how you would evaluate answer grounding.

When would you use an LLM or an agentic workflow in support automation?

What to coverThe team builds LLM-based applications and agentic workflows for customer support. Suggested preparation: distinguish a direct model call from a multi-step workflow, and discuss reliability, oversight, and when a simpler approach would suffice.

How would you approach recommendations for a support workflow?

What to coverRecommendation systems and AI-powered recommendations are named in the posting. Suggested preparation: describe possible inputs and baselines, then explain how you would assess relevance and handle limited information about a new case.

Where might a knowledge graph help with case analysis or retrieval?

What to coverKnowledge graphs are listed among techniques to apply where relevant. Suggested preparation: describe a plausible relationship-based use case, what data it would require, and why a graph might or might not improve on simpler retrieval.

How would you test whether an AI service improves customer support?

What to coverThe role includes experimentation and delivering services integrated into support processes. Suggested preparation: distinguish offline model metrics from workflow outcomes, and explain how you would compare against a baseline without confusing correlation with improvement.

What would you address when moving a model from prototype to production?

What to coverThe posting calls for collaboration with engineers on production models, data and model pipelines, and ML microservices. Suggested preparation: cover reproducibility, interfaces, monitoring, failure handling, and trade-offs between model quality and service reliability.

How would you investigate slow or poor-quality retrieval at scale?

What to coverThe role involves large-scale data and retrieval systems; familiarity with tools such as Spark or Elasticsearch is described as valuable, not required. Suggested preparation: separate data, indexing, retrieval-quality, and latency causes, and explain how measurements would guide changes before choosing a tool.

Behavioural practice

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

What to coverThe posting emphasizes translating business and technical requirements with stakeholders. Suggested preparation: give a real situation, the questions and actions you took to align expectations, and the resulting decision or outcome.

Describe a time you worked with engineers to deliver a model or data product.

What to coverThe role involves collaboration with data engineers, machine learning engineers, and developers to deliver production-ready solutions. Suggested preparation: use a real situation/action/result example that shows your contribution, how responsibilities were coordinated, and what happened at handoff.

Tell me about a difficult data or model problem you solved independently.

What to coverIndependent problem-solving is an explicit expectation. Suggested preparation: describe the situation, how you diagnosed and prioritized the problem, when you sought input, and the result.

Describe a time you explained a model trade-off to a nontechnical stakeholder.

What to coverThe posting calls for communication with both business and technical stakeholders. Suggested preparation: give a real situation/action/result example that shows how you made the trade-off understandable and what decision followed.

Tell me about a time curiosity changed your approach to an AI problem.

What to coverCuriosity is named as an essential quality, alongside practical experimentation. Suggested preparation: describe a real observation that prompted investigation, the action you took to test an alternative, and the outcome—including an inconclusive result if that is what occurred.

Sharpen your knowledge

Topics to prepare

Support Delivery problem framing
The role translates business and technical requirements into scalable AI solutions for customer support. Suggested exercise: take a hypothetical case-analysis request and define the user, desired action, data needed, and a measurable outcome; discuss trade-offs with both business and technical stakeholders.
Model evaluation and experimentation
The posting explicitly calls for model development, evaluation, and experimentation. Suggested exercise: compare a simple baseline with a revised classification approach, choosing metrics and an error-analysis method that reflect the support use case.
Retrieval and LLM application design
The role names knowledge retrieval, LLMs, embeddings, RAG, and semantic search. Suggested exercise: sketch how a support question retrieves relevant knowledge and produces an answer, then explain how to assess retrieval relevance and unsupported answers.
Recommendations and automation boundaries
Recommendations, classification, and automation are listed support use cases. Suggested exercise: outline when a system should recommend a next action, classify a case, or defer a decision, and how you would evaluate each outcome.
Moving AI solutions into production
The role takes models from experiments to reliable solutions and designs data pipelines, model pipelines, and ML microservices where required. Suggested exercise: trace a model from input data through deployment and monitoring, identifying handoffs with engineers and likely failure points.

Skills to revise

Python analysis of mixed data
Strong Python and analysis of structured or unstructured data are explicit. Suggested exercise: use a small sample of hypothetical support records and text to inspect missing values, summarize patterns, and explain which findings could inform case analysis.
Feature engineering and model development
These are explicit practical data science requirements. Suggested exercise: build a baseline case classifier on sample data, add one defensible feature, and compare results while checking for data leakage.
Retrieval prototype and assessment
The role applies embeddings, RAG, and semantic search where relevant. Suggested exercise: prototype retrieval over sample knowledge articles and review the top results for a set of support questions; state when retrieval does not provide enough evidence for an answer.
Pipeline and service implementation planning
Designing and implementing pipelines and ML microservices is part of the role where required. Suggested exercise: draft inputs, outputs, validation, and failure handling for a case-classification service and its model pipeline; identify what to coordinate with ML engineers and developers.
Stakeholder explanation
Communication and collaboration with business stakeholders and engineering teams are explicit. Suggested exercise: present a proposed support AI solution in two brief versions—one focused on business value and limitations, the other on data, evaluation, and production dependencies.

Your preparation checklist

Map the role to your evidence
Create a one-page inventory of evidence you can substantiate for the stated degree, 3+ years of hands-on ML or applied AI experience, Python, and collaboration with engineers. Leave gaps as questions, not claims.
Prepare an end-to-end project walkthrough
Select a relevant project you can discuss and outline the problem, data, feature engineering, model choices, experiments, evaluation, results, and your individual contribution.
Practice a customer-support design exercise
Sketch a proposed solution for case classification, knowledge retrieval, or recommendations. Explain how you would clarify business and technical requirements, handle structured and unstructured data, and define success measures.
Compare AI approaches and evaluation plans
Prepare a short comparison of when embeddings, semantic search, RAG, or an LLM might help a support use case, including baselines and ways to test retrieval quality and answer usefulness. These are preparation examples, not prescribed SAP methods.
Explain a path from prototype to production
Draw a simple data and model pipeline or ML microservice design, noting how you would work with data engineers, ML engineers, and developers to make it reliable. The posting says these components are built where required.
Prepare a stakeholder communication example
Rehearse how you translated an ambiguous request into a measurable AI task, communicated trade-offs to business and technical stakeholders, and incorporated feedback; use an example you can substantiate.
Bring targeted questions for the team
Ask which support use cases are priorities, how the team evaluates production-ready solutions, and where data scientists hand off or share ownership with engineering. These are questions to investigate, not stated team processes.

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 Singapore. Matched to this role, not your profile.

Explore more jobs

Finding similar opportunities…