New opportunity

Machine Learning Scientist

DP World · Bangalore, India

About this role

The Machine Learning Scientist develops ML solutions for planning, sequencing, routing, allocation, and resource-utilization decisions. The position is based in Bangalore and is listed as full-time.

Responsibilities

Formulate real-world decision problems and build and evaluate solutions using optimization and solvers, including MILP, CP-SAT, constraint programming, heuristics and search; deep reinforcement learning and decision intelligence; and predictive models that inform decisions.
Prototype rapidly with agentic coding assistants while maintaining code quality through testing, refactoring and disciplined experimentation.
Build evaluation harnesses using offline simulation, counterfactual testing, ablations and scenario analysis. Define KPIs and acceptance thresholds.
Work with ML engineers on production readiness, including latency, throughput, monitoring, reproducibility, model versioning and safe rollout. Document technical work and communicate findings to technical and non-technical stakeholders.

Required:

0–5 years of applied ML, data science or applied research experience; internships, thesis work and strong project portfolios count. The job information lists a bachelor's degree level.
Demonstrated use of agentic coding assistants in real development, strong Python skills, familiarity with ML tooling, and foundations in algorithms, probability, statistics and experimental design. PyTorch is preferred; TensorFlow is acceptable.
Ability to turn messy problems into clear formulations and measurable success metrics.

Preferred

Deep RL experience, including PPO, SAC, DQN, offline RL, imitation learning or MCTS/planning hybrids; simulator development, reward design, debugging and evaluation are relevant.
Simulation-based evaluation or digital twins; MLOps basics such as MLflow, Docker, CI/CD and model monitoring; cloud platforms; and exposure to logistics, supply chain or industrial operations. Indicative tools also include OR-Tools or other solvers, Ray RLlib or Stable Baselines, SQL and Git.

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

Skills for this role

PythonMachine LearningOptimizationMixed-Integer Linear ProgrammingCP-SATConstraint ProgrammingHeuristicsMetaheuristicsReinforcement LearningOffline Reinforcement LearningMulti-Armed BanditsMonte Carlo Tree SearchForecastingSimulationCounterfactual EvaluationExperimental DesignProbabilityStatisticsAgentic Coding AssistantsClaude CodePyTorchTensorFlowOR-ToolsRay RLlibStable BaselinesSQLDockerGitMLflowCI/CD

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

Formulate operational decisions
The role centers on planning, sequencing, routing, allocation and resource utilization. It requires turning messy problems into clear formulations and measurable success metrics.
Build decision-making approaches
The job calls for developing and evaluating optimization and solver methods, deep RL or decision-intelligence approaches, and predictive models that feed decisions. Prepare to explain why a method fits a particular decision problem.
Evaluate against defined thresholds
The role explicitly includes offline simulation, counterfactual testing, ablations and scenario analysis, with KPIs and acceptance thresholds defined in advance.
Prototype quickly without losing rigor
Demonstrated use of agentic coding assistants in real development is required. The work pairs fast scaffolding and experimentation with tests, refactoring and code quality.
Support production and communicate results
The scientist collaborates with ML engineers on latency, throughput, monitoring, reproducibility, versioning and safe rollout, then documents findings for technical and non-technical stakeholders.

Likely challenges to think through

Balancing solution quality and runtime
Preparation hypothesis: routing or allocation formulations may need to trade solution quality against solver or inference time. Prepare to explain how to compare a feasible baseline with more complex methods under a stated time limit.
Making offline results credible
Preparation hypothesis: a policy that performs well in simulation may not hold up under different operational scenarios or imperfect historical data. Prepare to discuss counterfactual limitations, ablations and scenario tests before setting an acceptance threshold.
Keeping rapid prototypes production-ready
Preparation hypothesis: agent-assisted iteration can increase speed while making experiments harder to reproduce or maintain. Prepare to describe how tests, versioning and documented experiments would support a handoff to ML engineers.

Practice your answers

Technical practice

How would you turn a messy routing or allocation request into a decision problem?

What to coverSuggested preparation: Explain how you would identify decisions, constraints, available data and an objective before choosing a method. Describe a measurable success metric and how you would check that the formulation reflects the operational problem.

When would you use MILP, CP-SAT or constraint programming for a sequencing problem?

What to coverSuggested preparation: Compare how each approach represents constraints and objectives, and discuss solution quality versus runtime. Explain how you would validate feasibility and handle a solver that cannot find an acceptable solution within the available time.

How would you design a heuristic baseline for a planning problem?

What to coverSuggested preparation: Outline a simple feasible baseline and how search or a metaheuristic might improve it. Compare its runtime, robustness and solution quality with a solver-based approach.

When might reinforcement learning be preferable to optimization for repeated decisions?

What to coverSuggested preparation: Discuss the need for sequential feedback, an environment, a reward definition and a credible baseline. Deep RL is preferred rather than required in the posting; if relevant to your experience, explain how you would evaluate policy learning against simpler approaches.

How would you assess an offline RL policy before considering its use?

What to coverSuggested preparation: Explain why historical decisions may not cover the actions a new policy would take, and how that limits evaluation. Discuss suitable baselines, counterfactual tests and safeguards against overinterpreting offline results.

How could a forecasting model improve an allocation decision?

What to coverSuggested preparation: Describe what the model would estimate and how its predictions would enter the decision system. Address forecast error and compare end-to-end decision outcomes, not just predictive accuracy.

How would you build a simulation-based evaluation harness?

What to coverSuggested preparation: Explain how you would represent key constraints and variability, then test decisions across scenarios. Discuss simulator validation and why simulated gains may not transfer to real operations.

How would you define acceptance thresholds and use ablations to evaluate a new approach?

What to coverSuggested preparation: Choose KPIs tied to the decision objective and identify constraints that must not regress. Explain how ablations and experimental design would distinguish a genuine improvement from noise or an added component that contributes little.

How would you use an agentic coding assistant without sacrificing code quality?

What to coverSuggested preparation: Describe where generated scaffolds, refactoring or test drafts could accelerate a Python prototype. Explain how you would review changes, test behavior and keep experiments reproducible.

What would you check before handing a decision model to ML engineers for productionization?

What to coverSuggested preparation: Discuss latency, throughput, monitoring, reproducibility and model versioning in relation to the proposed solution. Explain how you would agree on safe rollout checks; the posting names these concerns but does not prescribe a deployment process.

Behavioural practice

Tell me about a time you clarified an ambiguous analytical problem.

What to coverSuggested preparation: Use a real work, internship, thesis or project example. Describe the situation, the questions you asked, the formulation or metrics you chose, and the resulting decision or learning.

Describe a time an experiment or model failed to meet its goal.

What to coverSuggested preparation: Explain the situation, how you diagnosed the issue and what you changed or ruled out. Close with the measured result or what the failure taught you, without claiming an improvement you cannot substantiate.

Tell me about a time you used an agentic coding assistant on a real development task.

What to coverSuggested preparation: Describe the task, what you delegated to the assistant and how you checked its output through review or tests. State the result, including any correction or limitation you encountered.

Describe a time you worked with an engineer or another collaborator to make a model usable.

What to coverSuggested preparation: Identify the situation, your contribution and how you addressed an implementation constraint or handoff concern. Explain the outcome; a project collaboration is a valid example if it is your real experience.

Tell me about a time you explained technical findings to a non-technical audience.

What to coverSuggested preparation: Describe the audience's decision, how you communicated results and uncertainty, and what action followed. Keep the example grounded in what you actually did and observed.

Sharpen your knowledge

Topics to prepare

Decision-problem formulation
Required: the role turns messy planning, sequencing, routing, allocation and resource-utilization problems into formulations and measurable success metrics. Revise by writing the decisions, constraints, objective and two KPIs for a small allocation problem.
Optimization and search methods
The posting names MILP, CP-SAT, constraint programming, heuristics and search. Revise by comparing how each would represent a routing or scheduling constraint and when an approximate solution may be useful.
Probability, statistics and experimental design
These are required foundations. Revise by designing a comparison between two decision policies, specifying a baseline, sources of variation and how you would assess whether a KPI change is meaningful.
Decision-system evaluation
The role calls for offline simulation, counterfactual testing, ablations, scenario analysis and acceptance thresholds. Revise by outlining what each would reveal—and what it could not establish—for a proposed allocation policy.
Deep reinforcement learning and planning
Deep RL is a preferred differentiator, not a requirement. If preparing for it, revise reward design, offline RL and MCTS-style planning by explaining how you would evaluate a learned policy against a simple baseline.
Predictive models in decision systems
The posting includes forecasting and estimation models that feed decisions. Revise by tracing how forecast error could change a downstream resource-allocation decision and its KPI.

Skills to revise

Prototype and test with an agentic coding assistant
Required: demonstrated use of agentic coding assistants in real development without sacrificing code quality. Practise using one to scaffold a small Python decision-system prototype, then review its output, add tests, refactor and record experimental changes.
Implement a constrained optimization baseline
The role develops solver-based solutions; OR-Tools is listed as indicative, not required. Practise coding a small assignment or sequencing problem in Python, enforcing constraints and comparing its result with a simple heuristic.
Build an offline evaluation harness
The posting explicitly asks for simulation, counterfactual testing, ablations and scenario analysis. Practise running two policies over the same simulated scenarios, reporting KPIs and checking a stated acceptance threshold.
Train and inspect a predictive model
Strong Python and comfort with ML tooling are required; PyTorch is preferred and TensorFlow is acceptable. Practise training a small forecasting or estimation baseline and inspecting how its errors affect a subsequent decision.
Explain production trade-offs and findings
The role involves latency, throughput, monitoring, reproducibility, model versioning, safe rollout and communication. Practise producing a short technical note that reports an experiment, its limitations and the production checks you would discuss with ML engineers.

Your preparation checklist

Prepare a decision-problem walkthrough
Choose a real project or a clearly labeled hypothetical planning, routing, sequencing, or allocation problem. Write a one-page formulation covering decisions, constraints, uncertain inputs, and measurable success metrics.
Practice an optimization design exercise
Sketch a MILP or CP-SAT formulation and a heuristic baseline for that problem. Be ready to explain feasibility, trade-offs, and how you would compare the approaches.
Select evidence of applied ML and Python work
Prepare a concise walkthrough of relevant internship, thesis, or portfolio work, including your contribution, Python and ML tools used, experiments, results, and limitations.
Document an agentic coding workflow
Show a real development example, if available, where an agentic assistant helped scaffold, refactor, test, or iterate. Identify what you verified yourself and how you maintained code quality.
Draft an evaluation plan
Outline an offline simulation or scenario-analysis harness with baselines, ablations, KPIs, and proposed acceptance thresholds. State which assumptions would need validation against operational data.
Prepare to discuss RL and production trade-offs
If you have relevant RL work, explain the environment, reward design, baseline, debugging, and evaluation; deep RL is preferred, not required. Sketch how you would discuss latency, monitoring, reproducibility, versioning, and safe rollout with ML engineers.
Bring focused questions for DP World
Ask which decisions this role will support first, what data and simulators are available, how success is measured, and how scientists and ML engineers divide production responsibilities.

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…