New opportunity

Data Scientist

Celebal Technologies · Jaipur, India

About this role

Celebal Technologies seeks a Data Scientist for an on-site, permanent position in Jaipur, Rajasthan, India. The posting specifies 3–7 years of experience. The role develops and maintains AI-based systems, translating mathematical and business challenges into machine learning solutions across classical ML, deep learning, conversational AI and generative AI.

Responsibilities

Develop generative AI solutions and build or fine-tune models including GANs, VAEs and Transformers. Design classical ML models using methods such as regression, decision trees, SVMs, random forests, gradient boosting, clustering and dimensionality reduction.
Train, fine-tune and deploy deep learning models, including CNNs, RNNs and LSTMs. Work on NLP tasks and refine prompts to improve outcomes from LLMs such as GPT, BERT and T5.
Prepare large datasets through preprocessing, augmentation and feature engineering. Optimize hyperparameters using grid search, random search or Bayesian optimization; evaluate accuracy, precision, recall and F1-score; and monitor for drift, overfitting and bias.
Participate in code and design reviews, validate models with engineers, improve system quality and scalability, and collaborate with data scientists, engineers and product managers to bring ML solutions into production. Apply relevant AI/ML research and emerging techniques.

Qualifications

Bachelor’s or Master’s degree in Computer Science, Mathematics, Statistics, Data Science or a related field. Extensive experience in classical ML and deep learning, including frameworks such as TensorFlow, PyTorch or Keras.
Proficiency in Python and libraries including NumPy, Pandas and Scikit-learn; experience with NLP libraries such as Hugging Face Transformers, NLTK or spaCy. Strong knowledge of statistical modeling, probability and mathematical optimization.
Experience deploying ML models to production using AWS SageMaker, Azure ML, GCP AI or similar platforms, and experience with code review and design optimization in large-scale AI/ML systems.

Additional advantages include prompt engineering and LLM fine-tuning experience, familiarity with MLOps, and understanding of distributed computing and parallel processing. The role also calls for English communication, presentation skills, teamwork, self-motivation, attention to detail, ownership, reliability, willingness to learn and the ability to meet deadlines while handling high volumes of work.

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

Skills for this role

PythonMachine LearningDeep LearningGenerative AIConversational AINatural Language Processing (NLP)Large Language Models (LLMs)Prompt EngineeringModel Fine-tuningTensorFlowPyTorchKerasNumPyPandasScikit-learnHugging Face TransformersNLTKspaCyAWS SageMakerAzure Machine LearningGCP AIMLOpsFeature EngineeringData PreprocessingHyperparameter OptimizationModel DeploymentModel MonitoringStatistical ModelingProbability TheoryCode Review

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 problems into model choices
The role calls for turning complex mathematical and business challenges into actionable models, including classical methods such as regression, trees and clustering. Prepare to explain why a particular approach fits a problem rather than defaulting to the most complex model.
Build and refine deep-learning and generative systems
The posting includes training, fine-tuning and deploying deep-learning models; developing generative AI solutions; and improving NLP and LLM outcomes through prompt optimization. Prepare a clear account of model selection, iteration and observed results.
Make data and evaluation dependable
Responsibilities span large-dataset preprocessing, augmentation, feature engineering, hyperparameter optimization and evaluation using accuracy, precision, recall and F1. The role also monitors drift, overfitting and bias. Prepare to explain how data decisions and metric choices affect model conclusions.
Bring models into production with others
The role involves production deployment, code and design reviews, model validation with engineers, and collaboration with data scientists and product managers. Prepare to discuss decisions that improved system quality or scalability and how you communicated them across functions.

Likely challenges to think through

Breadth versus depth
Preparation hypothesis, not a confirmed employer problem: the range from classical ML to generative AI could make it difficult to choose where added complexity is justified. Prepare a comparison showing when a simpler baseline is preferable and what evidence would warrant a more complex model.
Offline gains versus sustained performance
Preparation hypothesis: improving a reported metric may conflict with guarding against overfitting, bias or later drift. Prepare to explain validation choices, metric trade-offs and what you would monitor after deployment.
Experimentation versus production quality
Preparation hypothesis: applying emerging techniques while meeting review, scalability and deadline expectations may require limiting experimentation. Prepare an example of how you would test an idea, decide whether to adopt it and make the resulting system reviewable.

Practice your answers

Technical practice

How would you choose a classical ML approach for a business problem?

What to coverSuggested preparation: Compare regression, tree-based methods, SVMs, clustering and dimensionality reduction according to the target, data and need for interpretability. Explain how statistical assumptions and validation would inform your choice.

How would you prepare a large dataset for modeling?

What to coverSuggested preparation: Discuss preprocessing, feature engineering and when augmentation is appropriate. Explain how you would use Python tools such as Pandas and NumPy while preventing information from the validation data from leaking into training.

When would you use a CNN, RNN, LSTM or Transformer?

What to coverSuggested preparation: Relate each architecture to the structure of the data and the problem it solves. Discuss training and fine-tuning trade-offs, including computational cost and overfitting.

How would you evaluate a GAN, VAE or Transformer for a generative AI task?

What to coverSuggested preparation: Explain what each model family is suited to generating and how you would assess the usefulness and quality of its outputs. Address trade-offs between output quality, training complexity and the intended application.

How would you improve an LLM's responses through prompt refinement?

What to coverSuggested preparation: Describe a repeatable way to revise prompts and compare outcomes on representative NLP tasks. Explain how you would check whether an apparent improvement holds beyond a few examples.

When would you fine-tune an NLP model rather than refine its prompts?

What to coverSuggested preparation: Compare the data, effort and evaluation needed for fine-tuning with those needed for prompt changes. Ground the discussion in models named in the posting, such as GPT, BERT or T5, without treating them as interchangeable.

How would you select metrics and tune hyperparameters?

What to coverSuggested preparation: Explain when accuracy, precision, recall or F1-score best reflects the problem, and what each can obscure. Compare grid, random and Bayesian search, including how you would keep evaluation separate from tuning.

How would you investigate declining model performance after deployment?

What to coverSuggested preparation: Distinguish data or model drift from overfitting and examine whether performance differs across relevant groups. Describe what evidence would prompt further validation, retraining or a change to the model.

How would you take a validated ML model into production?

What to coverSuggested preparation: Outline the handoff from model validation to deployment and monitoring, using one of the posting's example platforms—AWS SageMaker, Azure ML or GCP AI—if familiar. Discuss reliability and the trade-off between model performance and operational complexity.

What would you examine in a review of a large-scale ML system?

What to coverSuggested preparation: Cover code quality, reproducible validation and design choices affecting system efficiency and scalability. Explain when distributed or parallel processing might help; the posting identifies knowledge of these as a plus, not a required tool.

Behavioural practice

Tell me about collaborating to bring an ML solution toward production.

What to coverSuggested preparation: Use a real situation involving different technical or product perspectives, then describe your actions and the result. Make clear how you resolved a concrete disagreement or dependency.

Describe a time a code or design review changed your approach.

What to coverSuggested preparation: State the situation, the feedback you gave or received, the action taken and the result. Focus on how the review improved quality, validation or scalability rather than merely noting that a review occurred.

Give an example of presenting a complex model decision clearly.

What to coverSuggested preparation: Identify your audience and the decision they needed to make, then explain how you communicated evidence and trade-offs. Close with the result and how you checked that the explanation was understood.

Tell me about maintaining accuracy under a heavy workload and deadline.

What to coverSuggested preparation: Describe the situation, how you prioritized and checked your work, and the actual outcome. Include how you communicated risks or changes in timing rather than assuming the deadline was met.

Describe a time you learned and applied a new AI or ML technique.

What to coverSuggested preparation: Explain why the technique was worth considering, what you did to assess it, and the result. Distinguish evidence of an improvement from enthusiasm for a newer approach.

Sharpen your knowledge

Topics to prepare

Classical ML and statistical reasoning
The posting names regression, trees, SVMs, ensembles, clustering and dimensionality reduction, alongside probability and optimization. Suggested exercise: compare two methods for a hypothetical business problem and explain their assumptions, trade-offs and validation approach.
Deep learning and generative model architectures
The role covers CNNs, RNNs, LSTMs, GANs, VAEs and Transformers. Suggested exercise: sketch how three of these architectures learn, when each fits a problem, and what fine-tuning would change.
NLP, LLMs and conversational AI
The posting calls for NLP experience and work refining prompts for models such as GPT, BERT and T5. Suggested exercise: distinguish their relevant uses, then explain how you would evaluate a prompt change against a defined NLP task.
Model evaluation and failure modes
The role specifies accuracy, precision, recall, F1-score, hyperparameter optimization, drift, overfitting and bias. Suggested exercise: interpret a small confusion matrix, choose a metric, and explain how you would detect each failure mode.
Production ML system design
The posting calls for deployment, scalable design, code review and collaboration with engineers and product managers. Suggested exercise: outline a model's path from validation to production and identify review points and performance checks.

Skills to revise

Prepare data with Python
Python, NumPy, Pandas, preprocessing, augmentation and feature engineering are named in the posting. Suggested exercise: clean a sample dataset, document missing-value decisions and produce features without leaking validation data.
Run a classical ML comparison
The role includes classical models and grid, random or Bayesian hyperparameter search. Suggested exercise: use Scikit-learn to compare a regression or classification baseline with a second named method, tune parameters and report the relevant metrics.
Train and assess a deep learning model
The posting asks for training and fine-tuning with frameworks such as PyTorch, TensorFlow or Keras. Suggested exercise: train a small model in one of those frameworks, inspect training versus validation results, and explain a performance adjustment.
Test an NLP prompt or fine-tuning change
NLP libraries including Hugging Face Transformers, NLTK and spaCy are named; prompt engineering and LLM fine-tuning are described as a strong plus. Suggested exercise: hold a small set of inputs constant while comparing two prompts or model settings, and record outcome differences.
Explain a deployment and monitoring workflow
Production deployment using AWS SageMaker, Azure ML, GCP AI or similar platforms is specified; MLOps familiarity is an advantage. Suggested exercise: walk through deploying a trained model on one named platform and checking its performance for drift after release.
Practice code and design review
The posting explicitly calls for code review, model validation with engineers and scalable system design. Suggested exercise: review a short Python model pipeline for data leakage, reproducibility and inefficient processing, then present prioritized fixes.

Your preparation checklist

Map evidence to the posting
The posting specifies 3–7 years of experience and asks for a related degree, Python, classical ML, deep learning and production deployment. Prepare a one-page map of relevant experience and examples you can substantiate; leave gaps explicit.
Prepare a classical ML project walkthrough
Choose a project you can discuss end to end. Be ready to explain the business problem, preprocessing and feature engineering, why you selected a method, the statistical reasoning, and results using appropriate metrics.
Show how you evaluated and improved a model
Prepare a compact comparison of baselines and experiments, including any grid, random or Bayesian hyperparameter search you used. Explain validation choices, overfitting checks and what precision, recall or F1 revealed where relevant.
Walk through an NLP or generative AI example
The role covers NLP, LLM prompt optimization and model fine-tuning. If you have a relevant example, prepare prompts or model choices, evaluation results, failure cases and the changes you made; distinguish work you performed from work you only studied.
Rehearse a production design exercise
Sketch how you would take an ML model from prepared data to deployment on a suitable platform, then monitor performance, drift and bias. Treat the architecture as a proposed approach, not as Celebal Technologies’ existing system.
Prepare a collaboration and review example
The posting calls for code and design reviews and work with engineers, data scientists and product managers. Prepare one concise account of a review or cross-functional decision, your contribution, the trade-off and the outcome.
Write targeted questions for the employer
Ask which classical ML, deep learning or conversational AI problems this team currently prioritizes; how models are validated and monitored; and what deployment platforms and review practices the team uses.

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…