New opportunity

Data Scientist

Viridien · Airoli, India

About this role

Viridien is hiring a remote Data Scientist for its Airoli, India location. The role focuses on preparing and integrating data, improving annotation workflows, and applying machine learning within the Data Hub. Viridien works on Earth data and technology challenges involving natural resources, energy transition, digital systems, infrastructure and geoscience.

Responsibilities

Develop systematic modules to gather, clean, validate, transform and structure raw datasets. Build scalable, reusable data-processing solutions and maintain version control in GitLab.
Work with subject matter experts and the labeling team to understand domain requirements, provide tools and technical feedback, and improve annotation workflows. Diagnose data problems and flag inconsistencies or risks.
Maintain knowledge of Data Hub systems, schemas, emerging technologies, models and research. Contribute to data integration, feature design and machine-learning initiatives.
Design and run experiments, iterate on results, and document and share findings. Use pre-trained ML models and train or fine-tune models on internal data with appropriate evaluation and validation.
Communicate with the team, its lead, and production and technology partners. The role is assessed on timely, high-quality delivery, learning and applying new concepts, and understanding Data Hub workflows.

Qualifications

Background in data science or a related field; a master’s degree is preferred.
Proficiency in at least one programming language, ideally Python, and experience with common ML libraries and hybrid workflows involving traditional ML, LLMs, embeddings, ontologies and knowledge graphs.
Ability to work with relational, NoSQL and graph databases, plus strong data-processing skills in cleaning, filtering and feature extraction. Clear communication, collaboration and presentation skills are required.

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

Skills for this role

PythonMachine LearningLarge Language ModelsEmbeddingsOntologiesKnowledge GraphsRelational DatabasesNoSQL DatabasesGraph DatabasesData CleaningData ValidationData TransformationFeature ExtractionData IntegrationModel Fine-TuningModel EvaluationGitLabExperimentationCommunicationCollaborationPresentation

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

Reusable data preparation
The role develops modules to gather, clean, validate and structure raw datasets, with scalable processing and GitLab version control. Prepare to explain how you would make transformations reproducible and detect data issues early.
Annotation workflow improvement
The posting calls for close work with subject matter experts and the labeling team, including tools and technical feedback. Prepare an example of turning domain requirements or labeling feedback into a workflow improvement.
Data Hub integration and feature design
The role requires understanding Data Hub systems and schemas and contributing to data integration and feature design. Prepare to discuss data modeling across relational, NoSQL and graph databases without assuming a particular architecture.
Validated ML experimentation
The posting calls for experiments, documented iterations, and appropriate evaluation of pre-trained or fine-tuned models on internal data. Prepare to explain baseline selection, validation and how results would guide the next iteration.
Timely, transparent delivery
The stated performance measures include high-quality work on time, rapid learning and understanding Data Hub workflows. Regular communication with the team lead and production and technology partners is also expected; prepare to explain how you report progress and risks.

Likely challenges to think through

Data quality versus delivery speed
A likely trade-off, not a confirmed problem, is how much cleaning and validation to do before downstream work can proceed. Prepare to explain checks that catch consequential inconsistencies, when to flag a risk and how to document unresolved limitations.
Labeling consistency versus domain nuance
Working with experts and labelers may require reconciling ambiguous cases without making annotation tools too rigid. Prepare an approach for collecting examples, clarifying guidance with experts and checking whether changes improve consistency.
Model complexity versus evidence of value
With traditional ML, LLMs, embeddings, ontologies and knowledge graphs in scope, a plausible challenge is choosing an approach that adds value without unnecessary complexity. Prepare to compare a simple baseline with a more complex option using appropriate evaluation and validation.

Practice your answers

Technical practice

How would you design a module to turn raw datasets into structured data?

What to coverSuggested answer focus: Walk through gathering, cleaning, validation, transformation and a defined output structure. Explain how you would make the steps systematic and test the resulting data.

How would you diagnose an inconsistency in data flowing through the Data Hub?

What to coverSuggested answer focus: Trace the issue through source data, transformations and the relevant schema to locate its cause. Explain how you would assess the risk, document evidence and flag it to the appropriate collaborators.

How would you make a data-processing solution reusable and scalable?

What to coverSuggested answer focus: Discuss modular interfaces, repeatable validation and how you would check behaviour as data volumes or sources change. Explain how GitLab version control would help the team review and maintain changes.

How would you improve an annotation workflow with SMEs and a labeling team?

What to coverSuggested answer focus: First establish the domain requirements and where annotators encounter ambiguity or repeated work. Then describe how you would test a tool or process change and use technical feedback to refine it.

How would you integrate data across relational, NoSQL and graph databases?

What to coverSuggested answer focus: Compare the structures each database supports and explain how schemas and relationships affect integration. Describe checks you would use to preserve meaning and consistency when moving or combining data.

How would you design features from cleaned data for an ML initiative?

What to coverSuggested answer focus: Connect feature choices to the task and available domain knowledge, then explain how you would check data quality and usefulness. Discuss the trade-off between informative features and features that are difficult to maintain or validate.

When would you combine traditional ML with LLMs, embeddings, ontologies or knowledge graphs?

What to coverSuggested answer focus: Compare what each approach contributes to a task and where a simpler approach may suffice. Describe how you would test whether a hybrid workflow adds value rather than assuming it does.

How would you decide between using a pre-trained model and fine-tuning it on internal data?

What to coverSuggested answer focus: Consider task fit, the amount and quality of internal data, and the cost of training and maintenance. Explain what evidence you would seek before choosing an approach.

How would you evaluate and validate a model trained on internal data?

What to coverSuggested answer focus: Define success measures that match the task, use an appropriate separation of training and evaluation data, and inspect errors. Explain how you would detect whether an apparent improvement generalizes.

How would you run and document an experiment after an initial ML result disappoints?

What to coverSuggested answer focus: State a hypothesis, change a defined aspect of the data or model, and compare results against the initial run. Describe what you would record and share so the team can understand the outcome and next step.

Behavioural practice

Tell me about a time you clarified domain requirements with subject matter experts.

What to coverSuggested preparation: Use a real example to set out the situation, the questions you asked and how you translated the answers into data or labeling work. Close with the result, including any remaining uncertainty.

Tell me about a time you identified a data inconsistency or risk.

What to coverSuggested preparation: Describe how you noticed and investigated it, whom you informed, and what action followed. State the result without overstating what you could verify.

Tell me about a time you had to learn a new tool or concept to complete work.

What to coverSuggested preparation: Explain the task, how you learned and tested the concept, and how you applied it. Give the outcome and what you would do differently next time.

Tell me about a time you balanced a delivery deadline with data quality.

What to coverSuggested preparation: Describe the deadline and quality concern, the trade-off you communicated, and the action you took. Explain the result and how stakeholders understood any remaining risk.

Tell me about a time you shared experimental findings with cross-functional partners.

What to coverSuggested preparation: Set out the experiment and the audience, then explain how you communicated results and limitations. Describe the decision or next step that followed.

Sharpen your knowledge

Topics to prepare

Data quality and transformation design
Map how raw records would be gathered, cleaned, validated and structured; list likely failure modes and checks for each stage. The posting explicitly calls for systematic transformation modules and diagnosing inconsistencies.
Data Hub schemas and database models
Sketch how the same small dataset could be represented in relational, NoSQL and graph databases, then explain how schema differences affect integration. The role requires comfort with all three and understanding Data Hub schemas; its actual schemas are not provided.
Annotation workflows and domain requirements
Outline questions for a subject matter expert about ambiguous labels, then describe how feedback would reach the labeling team. The role explicitly involves improving annotation workflows with SMEs and labelers.
Hybrid machine-learning approaches
Compare when traditional ML, LLMs, embeddings, ontologies and knowledge graphs might contribute to one data-integration problem. These approaches are named in the qualifications; any proposed combination is a preparation example, not a stated company design.
Experiment and model-validation principles
Draft an experiment plan specifying a baseline, evaluation measures, validation split and decision rule for iteration. The posting calls for experiments, pre-trained models, training or fine-tuning, and proper evaluation.

Skills to revise

Build a reusable data-processing module
In Python, or another proficient language, process a sample dataset through cleaning, filtering, validation and feature extraction; include checks for malformed records. This rehearses the stated data-processing responsibilities; Python is preferred, not mandatory.
Version changes in GitLab
Put the sample processing module under GitLab version control and record a change that fixes a data-quality issue. The posting explicitly names GitLab and reusable data-processing solutions.
Practice cross-database integration
Using a small illustrative dataset, retrieve or represent records from relational, NoSQL and graph structures and reconcile their identifiers into one consistent view. This is a suggested exercise for the stated database and integration work, not a claim about Data Hub infrastructure.
Improve an annotation handoff
Write a short labeling guide for an ambiguous sample, flag inconsistent labels and propose a concrete feedback loop with an SME. This mirrors the stated collaboration and annotation-improvement duties.
Run and evaluate a model experiment
Apply a pre-trained model to sample data, compare it with a simple baseline, and document evaluation results; if practical, test a fine-tuned version. The role explicitly involves model use or fine-tuning, validation and iteration.
Present findings and risks
Summarize one exercise in a brief update covering results, data issues, next steps and decisions needed from SMEs or production partners. Clear communication, presentation and sharing experiment learnings are explicit expectations.

Your preparation checklist

Prepare a data-processing walkthrough
The role calls for reusable modules that gather, clean, validate, transform and structure raw data. Prepare a real project walkthrough, or a clearly labeled sample, showing inputs, validation checks, failure handling and how the code is organized in GitLab.
Map a data issue from detection to resolution
Prepare one example of diagnosing an inconsistency or risk: what evidence exposed it, how you investigated it, who needed to know and how you verified the fix.
Sketch an annotation-workflow improvement
The posting emphasizes work with subject matter experts and a labeling team. As a preparation exercise, sketch a tool or feedback loop that could surface ambiguous labels, capture domain guidance and measure whether annotation quality improves.
Practice a Data Hub integration design
Prepare a short, hypothetical design for integrating datasets and extracting features across relational, NoSQL and graph data. State your schema assumptions and trade-offs; the posting does not specify Viridien’s internal schemas.
Document a model experiment
Prepare an experiment brief comparing a baseline with a pre-trained or fine-tuned model, including dataset splits, evaluation measures, error analysis, iteration decisions and how findings would be shared. If relevant, explain where embeddings, LLMs or a knowledge graph fit.
Rehearse a cross-functional update
Practice a concise update for a team lead and production or technology partners: progress, evidence of quality, a data risk, its impact and the next decision needed. This reflects the role’s communication and timely-delivery expectations.
Bring focused Data Hub questions
Ask how the team’s datasets and schemas are organized, where annotation bottlenecks occur, how model validation is handled and what high-quality delivery looks like in the first projects.

Availability last checked: 5 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…