New opportunity

Data Scientist

Epergne Solutions · Singapore

About this role

Epergne Solutions is seeking a Data Scientist in Singapore to develop scalable AI and data-driven solutions for business problems. The role involves collaborating with business stakeholders, data engineers and technical teams to build predictive models, deploy machine learning solutions and support data science best practices.

Responsibilities

Analyze large structured and unstructured datasets; perform feature engineering, model development, evaluation and optimization.
Design, develop and deploy machine learning and AI solutions, including deep learning, statistical and predictive models. Build end-to-end ML pipelines using MLOps best practices.
Apply generative AI techniques, LLMs, embeddings, vector databases and AI frameworks where appropriate. Review code, maintain documentation and ensure solutions are scalable, reliable and of high quality.
Keep current with emerging AI, machine learning and data science technologies.

Required qualifications

A bachelor’s or master’s degree in Computer Science, Data Science, Statistics, Mathematics, Engineering or a related quantitative field, and a minimum of 3 years of experience in Data Science or Machine Learning.
Strong proficiency in Python, SQL, R and shell scripting; experience in machine learning, deep learning, statistical modelling, predictive analytics, MLOps and end-to-end model deployment on cloud platforms. AWS is preferred.
Experience with Transformer models, reinforcement learning and optimization techniques; knowledge of generative AI, LLMs, LangChain, LangGraph, embeddings and vector databases. Experience with ELT/ETL, data transformation and relational databases is also required.
Strong problem-solving, analytical and critical thinking skills, plus communication, collaboration and stakeholder management skills.

Preferred qualifications

Experience building production-grade AI/ML applications; familiarity with cloud-native AI services and deployment pipelines; experience with data visualization and model performance monitoring; and knowledge of software development best practices, version control and CI/CD for ML solutions.

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

Skills for this role

PythonSQLRShell ScriptingMachine LearningDeep LearningStatistical ModellingPredictive AnalyticsFeature EngineeringMLOpsCloud Model DeploymentAWSTransformer ModelsReinforcement LearningOptimizationGenerative AILarge Language ModelsLangChainLangGraphEmbeddingsVector DatabasesETLELTData TransformationRelational DatabasesData VisualizationModel Performance MonitoringVersion ControlCI/CDProblem-Solving and Analytical Thinking as needed

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

Deliver models for business problems
The role calls for designing, evaluating, optimizing and maintaining predictive, statistical and deep-learning models. Prepare to explain how a business objective shaped model selection and evaluation.
Build deployable ML pipelines
The posting emphasizes end-to-end ML pipelines, MLOps and cloud deployment, with AWS preferred. Prepare to discuss the path from training to deployment and how you would maintain reliability.
Work across complex data sources
The role involves large structured and unstructured datasets, feature engineering, and data engineering practices. Prepare to explain data transformation choices and their effect on model quality.
Collaborate and maintain solution quality
Working with business stakeholders, data engineers and technical teams is a stated duty, as are code review, documentation, scalability and reliability. Prepare to describe how technical decisions are communicated and reviewed.
Apply generative AI selectively
The posting calls for LLMs, embeddings, vector databases and AI frameworks where appropriate. Prepare to explain when these approaches fit a business problem rather than treating them as the default.

Likely challenges to think through

Balancing model performance with production constraints
A likely challenge is choosing between a more complex model and one that is easier to deploy and maintain; this is a preparation hypothesis, not a reported employer problem. Prepare a concrete example of the evaluation, scalability and reliability trade-offs you would consider.
Turning mixed data into dependable inputs
Large structured and unstructured datasets may make feature consistency and data quality difficult across an end-to-end pipeline. Prepare to explain how you would check transformations and detect input changes; the posting does not specify its datasets or current issues.
Aligning technical choices with stakeholder needs
Collaboration across business and technical teams may require trading off predictive performance, delivery effort and interpretability. Prepare to explain how you would establish a useful success measure and communicate the trade-off without assuming a particular company process.

Practice your answers

Technical practice

How would you turn a business problem into a predictive-modeling task?

What to coverSuggested preparation: Explain how you would define the target, establish a baseline and choose evaluation measures that reflect the business objective. Address how you would check whether the available data can support the proposed prediction.

How would you prepare large structured and unstructured datasets for modeling?

What to coverSuggested preparation: Discuss data quality checks, feature engineering and ways to handle missing or inconsistent information. Explain how you would prevent information leakage when creating training and evaluation datasets.

How would you use SQL and ELT or ETL to produce reliable model inputs?

What to coverSuggested preparation: Describe transformations and joins across relational data, including checks for duplicate records and unexpected row counts. Explain how you would validate the resulting dataset and make transformations reproducible.

When would you choose a statistical model over a deep-learning model?

What to coverSuggested preparation: Compare the approaches using data volume, interpretability, performance and operating cost. Explain how you would test whether added complexity provides a meaningful improvement.

How would you evaluate a Transformer or LLM for a business use case?

What to coverSuggested preparation: Explain how you would establish a simpler baseline and assess output quality, failure modes, latency and cost. Discuss when a generative approach would be inappropriate for the task.

How would you design retrieval using embeddings and a vector database?

What to coverSuggested preparation: Cover document preparation, embedding and retrieval choices, and evaluation of both retrieved material and final answers. If discussing LangChain or LangGraph, explain what the framework adds and when a simpler implementation would suffice.

When is reinforcement learning appropriate, and how would you evaluate an optimization approach?

What to coverSuggested preparation: Distinguish a sequential decision problem from one suited to conventional prediction or optimization. Explain how you would define the objective and constraints, then assess the approach against a simpler baseline.

How would you build and deploy an end-to-end ML pipeline on a cloud platform?

What to coverSuggested preparation: Walk through data preparation, training, evaluation, deployment and reproducibility using MLOps practices. Address reliability and scaling trade-offs; AWS is preferred in the posting, but no specific service is named.

How would you investigate a decline in production model performance?

What to coverSuggested preparation: Explain how monitoring could help distinguish data changes, pipeline faults and model degradation. Describe how you would verify the cause and decide whether to fix the pipeline, retrain or roll back.

How would you maintain the quality of a production AI or ML codebase?

What to coverSuggested preparation: Discuss code review, documentation, version control and CI/CD for model and pipeline changes. Explain how you would use Python, SQL, R or shell scripting where appropriate without assuming that every task needs every language.

Behavioural practice

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

What to coverSuggested preparation: Use a real situation to explain what was unclear, the questions you asked and the action you took to agree on a measurable goal. State the result and how you communicated any remaining limitations.

Describe a time you worked with data engineers or technical teams to deliver a data solution.

What to coverSuggested preparation: Give the situation, your responsibilities and the actions you took to coordinate data or deployment work. Explain the result, including how you resolved any handoff or ownership issue.

Tell me about a model or data analysis that did not meet expectations.

What to coverSuggested preparation: Describe the situation and the evidence that showed a problem, then explain how you diagnosed it and changed course. State the result without presenting an improvement that did not occur.

Describe a time you improved quality through code review or documentation.

What to coverSuggested preparation: Set out the quality concern, your specific action and how you worked with others to address it. Explain the result in terms of maintainability, reliability or clarity.

Tell me about a time you assessed whether to adopt a new AI or data-science technique.

What to coverSuggested preparation: Describe the situation, how you compared the technique with an existing approach and what evidence informed your recommendation. State the decision and result, including any trade-offs you communicated.

Sharpen your knowledge

Topics to prepare

Statistical and predictive modeling
The posting calls for statistical and predictive models, feature engineering, evaluation, and optimization. Suggested revision: explain how you would choose a baseline, prevent data leakage, select an evaluation metric, and decide whether a model improves on the baseline.
Deep learning and Transformers
Deep learning and experience with Transformer models are stated requirements. Suggested revision: sketch a Transformer’s main components and explain when you would compare it with a simpler predictive model.
Reinforcement learning and optimization
The posting explicitly names reinforcement learning and optimization techniques. Suggested revision: define states, actions, rewards, and an objective for a hypothetical decision problem, then explain how you would judge whether reinforcement learning is appropriate.
Generative AI architecture
The posting names LLMs, embeddings, vector databases, LangChain, and LangGraph, to be used where appropriate. Suggested revision: diagram how these components could support a retrieval-based application and identify quality and reliability checks.
Data engineering concepts
Large structured and unstructured datasets, ELT/ETL, transformations, and relational databases are in scope. Suggested revision: trace data from source through transformation to model-ready features, noting where data quality checks belong.
MLOps and deployment lifecycle
End-to-end ML pipelines, MLOps, and cloud model deployment are required; AWS is preferred. Suggested revision: outline the stages from training through deployment and monitoring, including how you would detect a decline in model performance.

Skills to revise

Python and R model development
Strong proficiency in both languages is requested. Suggested exercise: implement the same small predictive-model workflow in Python and R, documenting feature choices, evaluation results, and one optimization decision.
SQL and relational data preparation
SQL proficiency and experience with relational databases and transformations are requested. Suggested exercise: write queries that join, aggregate, and check source tables, then produce a model-ready dataset without leaking future information.
Shell scripting and pipeline operation
Shell scripting and end-to-end ML pipelines are stated requirements. Suggested exercise: write a shell script that runs data checks, training, and evaluation in sequence and stops clearly when a stage fails.
Cloud model deployment
The posting requests hands-on cloud deployment experience, with AWS preferred. Suggested exercise: rehearse explaining how you would package and deploy a model, verify predictions, and handle a failed release; do not assume a particular cloud service is required.
LLM application prototyping
The posting names LangChain, LangGraph, embeddings, and vector databases. Suggested exercise: sketch or prototype a small retrieval-based workflow using the relevant named components, then test whether retrieved context supports its answers.
Stakeholder communication and code quality
Collaboration with business stakeholders and technical teams, code review, and documentation are stated responsibilities; version control and CI/CD are preferred knowledge. Suggested exercise: prepare a concise model-results briefing for a business stakeholder and a technical handoff explaining assumptions, limitations, and review checks.

Your preparation checklist

Prepare a predictive-model project walkthrough
The role covers business problems, feature engineering, model evaluation and stakeholder collaboration. Prepare one real example showing the problem, data, modelling choices, results and how decisions were communicated; do not invent outcomes.
Build a data-to-features demonstration
Prepare a small example using structured or unstructured data to show SQL queries, transformation, feature creation and data-quality checks. Be ready to explain where Python, R or shell scripting fits; all four are listed as required skills.
Sketch an end-to-end deployment design
As a preparation exercise, draw a model pipeline from data ingestion through deployment and monitoring. Explain reliability, documentation, version control and CI/CD choices; the posting requires cloud deployment and prefers AWS.
Work through a generative-AI design exercise
Prepare a hypothetical use case and explain when an LLM, embeddings and a vector database would be appropriate, how you would evaluate the solution, and whether LangChain or LangGraph adds value. These technologies appear in the posting; no specific project is stated.
Rehearse modelling trade-offs
Prepare concise explanations of how you would choose and evaluate statistical, deep-learning or Transformer approaches, and when reinforcement learning or optimization would be suitable. Use examples you can substantiate rather than implying the role uses every method on each project.
Organize evidence for the stated qualifications
Create a one-page map of verifiable education, experience and work examples against the posting's degree, minimum three years of relevant experience, technical requirements and preferred production practices. Mark any gaps plainly rather than claiming experience you do not have.
Draft questions about the work
Ask which business problems and datasets are priorities, how data scientists work with stakeholders and data engineers, and what cloud deployment and model-monitoring practices are in place. The posting names these responsibilities but does not specify the team's current setup.

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…