New opportunity

AI/ML Engineer

Cittabase Solutions · Chennai, India

About this role

Cittabase Solutions seeks an AI/ML Engineer in Chennai to build scalable, production-grade AI applications. The role spans Generative AI, large language models (LLMs), NLP, predictive modeling and data-driven automation, working closely with data engineering and product teams.

Responsibilities

Design, train, deploy and optimize ML and Generative AI models for prediction, automation, NLP and decision intelligence.
Build LLM-powered chatbots, document intelligence systems, summarization engines and retrieval-augmented generation (RAG) pipelines using embeddings and vector search.
Perform data preprocessing, feature engineering, model evaluation and hyperparameter tuning. Manage training, validation, versioning, deployment and monitoring across end-to-end ML and GenAI pipelines.
Use Snowflake, Databricks and cloud platforms such as GCP, AWS and Azure to support scalable solutions. Monitor production performance, drift, cost and latency, and apply AI governance, security and responsible AI practices.
Collaborate across teams to turn business problems into AI solutions.

Qualifications

Bachelor’s or Master’s degree in Computer Science, Data Science, AI/ML or a related field, plus 2–3 years of hands-on ML/AI project experience.
Experience building LLM applications, including prompt engineering, RAG and integration with structured and unstructured data; experience with enterprise GenAI platforms and frameworks such as Snowflake Cortex, Databricks Mosaic AI and OpenAI-compatible APIs.
Strong Python skills and familiarity with XGBoost, PyTorch and TensorFlow; proficiency in SQL, data handling, feature engineering, model evaluation and deployment.
Experience with MLOps practices, including CI/CD for ML workflows and model versioning, and hands-on use of LangChain, LlamaIndex and LangGraph for agents and multi-step pipelines.

Preferred

Working experience with Snowflake and Databricks is a plus. The position is full-time; applications can be sent to careers@cittabase.com.

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

Skills for this role

Generative AILarge Language ModelsMachine LearningNatural Language ProcessingPredictive ModelingRetrieval-Augmented GenerationEmbeddingsVector SearchData PreprocessingFeature EngineeringModel EvaluationHyperparameter TuningPythonSQLXGBoostPyTorchTensorFlowPrompt EngineeringSnowflake CortexDatabricks Mosaic AIOpenAI-compatible APIsLangChainLlamaIndexLangGraphSnowflakeDatabricksGoogle Cloud PlatformAWSAzureMLOps CI/CD pipelines for ML workflows and model versioning are also required.

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

Build production-grade LLM applications
The role calls for chatbots, document intelligence, summarization and RAG pipelines using embeddings and vector search. Prepare to explain how you would design, evaluate and improve an LLM application that uses structured and unstructured data.
Deliver reliable predictive ML
Responsibilities include prediction, NLP and automation, supported by preprocessing, feature engineering, model evaluation and tuning. Prepare to explain how you would establish a baseline, select evaluation measures and test whether improvements hold up.
Own the deployment lifecycle
The posting covers training, validation, versioning, deployment and monitoring, including ML CI/CD. Prepare to describe a reproducible path from experiment to deployment and how you would respond to performance drift.
Use data and AI platforms effectively
The role involves SQL, Snowflake, Databricks and cloud platforms to support scalable solutions. Prepare to discuss data handling and integration choices, distinguishing hands-on experience from familiarity where relevant.
Translate business needs into governed solutions
The engineer works with data engineering and product teams and is expected to apply AI governance, security and responsible AI practices. Prepare to explain how you would define a useful outcome and account for data and model risks.

Likely challenges to think through

Balancing RAG quality against operating cost
A likely challenge, not a stated employer problem, is improving retrieval and answer quality without excessive latency or cost; the posting explicitly calls for monitoring both. Prepare to explain how you would evaluate retrieval and responses, then compare changes against latency and cost.
Keeping deployed models dependable
A likely trade-off is releasing improvements quickly while retaining validation, version history and drift visibility. Prepare to walk through checks before deployment, production monitoring and what would trigger investigation or rollback.
Choosing the right solution for a business problem
Because the role spans predictive ML, LLM applications and cross-functional work, a likely difficulty is deciding when a complex GenAI approach is justified. Prepare to compare a simpler baseline with an LLM-based option using expected value, evaluation results and governance risks.

Practice your answers

Technical practice

How would you design a RAG pipeline for a document intelligence application?

What to coverSuggested answer focus: Explain document preparation, embeddings, vector search, retrieval, and how retrieved context reaches the LLM. Discuss how you would assess answer quality and improve retrieval when relevant documents are missed.

How would you improve a chatbot or summarization engine that produces unreliable answers?

What to coverSuggested answer focus: Distinguish problems caused by prompts, retrieved context, and the model's output. Explain how you would evaluate changes for accuracy and reliability rather than relying on a few examples.

How would you integrate structured and unstructured data into an LLM application?

What to coverSuggested answer focus: Describe the different preparation and retrieval needs of SQL-accessible records and documents. Explain how you would preserve relevant context while checking that the data used is appropriate for the response.

How would you prepare features for a predictive modelling task?

What to coverSuggested answer focus: Cover data quality checks, preprocessing, SQL-based data handling, and feature engineering. Explain how you would prevent information leakage between training and validation data.

How would you evaluate and tune a predictive model?

What to coverSuggested answer focus: Explain how the business objective informs validation and metric choices, then describe a disciplined approach to hyperparameter tuning. Discuss accuracy, reliability, and scalability trade-offs when comparing models.

When would you use XGBoost rather than a PyTorch or TensorFlow model?

What to coverSuggested answer focus: Compare the approaches in terms of the data and prediction task, available training data, and deployment complexity. Explain how you would establish a baseline before choosing a more complex model.

How would you orchestrate a multi-step LLM workflow?

What to coverSuggested answer focus: Describe how components such as retrieval, model calls, and validation could be coordinated using LangChain, LlamaIndex, or LangGraph. Explain where an agent is useful and where a simpler, more predictable workflow may suffice.

How would you take an ML model from training to a versioned production deployment?

What to coverSuggested answer focus: Walk through validation, model versioning, deployment, and CI/CD checks for an ML workflow. Explain how you would make it possible to identify which model version is serving predictions.

How would you monitor and optimize a deployed AI application?

What to coverSuggested answer focus: Discuss model performance, drift, latency, and cost, which the posting identifies as production concerns. Explain how you would investigate a deterioration and weigh quality against operational constraints.

How would you address governance and security in an enterprise LLM application?

What to coverSuggested answer focus: Discuss controls over data access and model inputs and outputs, alongside evaluation for harmful or unreliable responses. Explain how those checks could fit into deployment and ongoing monitoring.

Behavioural practice

Tell me about a time you translated a business problem into an ML or AI solution.

What to coverSuggested answer focus: Use a real situation, your actions to clarify the objective with stakeholders, and the result. Explain how you decided what the system needed to do and how success would be judged.

Describe a time you worked with data engineering or product colleagues on a delivery.

What to coverSuggested answer focus: Give the situation, your specific contribution, and the result. Show how you resolved a data, integration, or requirements dependency across teams.

Tell me about a time a deployed model or application did not perform as expected.

What to coverSuggested answer focus: Describe the situation, the evidence you examined, the action you took, and the result. Be clear about how you distinguished a model-quality problem from a data or operational issue.

Describe a time you had to balance solution quality with latency or cost.

What to coverSuggested answer focus: Use a real situation, the options you considered, your decision, and its result. Explain what you measured and which trade-off mattered for the use case.

Tell me about a time you identified a security, governance, or responsible-AI concern.

What to coverSuggested answer focus: Describe the situation, the concern you identified, the action you took with relevant colleagues, and the result. Keep the example grounded in what you personally did rather than implying a policy or process at Cittabase.

Sharpen your knowledge

Topics to prepare

RAG and LLM application design
The posting calls for chatbots, document intelligence, summarization, embeddings, vector search and RAG. Suggested revision exercise: diagram a document-question-answering flow and explain how retrieval, prompting and evaluation affect answer quality.
Model evaluation and feature engineering
The role includes predictive modelling, preprocessing, feature engineering, evaluation and hyperparameter tuning. Suggested revision exercise: explain how you would split data, prevent leakage, choose metrics and compare a tuned model with a baseline.
Production ML and GenAI lifecycles
The posting specifies training, validation, versioning, deployment, monitoring and ML-workflow CI/CD. Suggested revision exercise: sketch a release flow and identify checks before deployment and signals that would trigger investigation afterward.
Agents and multi-step workflows
Hands-on experience with LangChain, LlamaIndex and LangGraph is listed for orchestrating agents and workflows. Suggested revision exercise: compare a fixed multi-step pipeline with an agent-driven flow, including when each would be appropriate.
Responsible operation at scale
The role mentions governance, security and responsible AI, alongside monitoring drift, cost and latency across Snowflake, Databricks and cloud platforms. Suggested revision exercise: outline safeguards and operational trade-offs for an LLM application using business data.

Skills to revise

Build a small RAG workflow
Suggested practice: ingest sample documents, create embeddings, retrieve relevant passages with vector search and generate a grounded response through an OpenAI-compatible API. Explain your prompt choices and how you would test retrieval and answers; these steps reflect the posted LLM requirements.
Prepare data and train a predictive model
Suggested practice: use Python and SQL to clean data and create features, then train and tune an XGBoost model. Report validation metrics and leakage checks. The posting explicitly calls for these data-handling and modelling skills.
Revisit neural-network implementation
Suggested practice: implement and evaluate a small NLP or prediction model in PyTorch or TensorFlow, then explain training choices and failure cases. The posting names both libraries and requires ML model development; using either one here is a revision suggestion, not a stated preference.
Orchestrate a multi-step LLM task
Suggested practice: build a short workflow with LangChain, LlamaIndex or LangGraph that retrieves information, produces a summary and handles a failed step. Be ready to explain the orchestration choice; the posting names these frameworks for multi-step AI pipelines.
Practice deployment and monitoring decisions
Suggested practice: outline model versioning and CI/CD checks for a deployment, then define how you would track performance, drift, latency and cost. The posting explicitly includes deployment, MLOps and production monitoring.

Your preparation checklist

Prepare a project walkthrough
Create a one-page account of an ML or GenAI project you actually worked on: business problem, your contribution, data, design decisions, evaluation and deployment outcome. The posting asks for 2–3 years of hands-on project experience.
Explain an LLM or RAG pipeline
Prepare a diagram and walkthrough covering prompts, structured and unstructured data, embeddings, vector search, retrieval and answer evaluation. Explain where you used LangChain, LlamaIndex or LangGraph, if applicable.
Show predictive-modeling evidence
Bring a concise example of Python and SQL data handling, feature engineering, validation and hyperparameter tuning. Compare a baseline with the final model and identify which listed libraries you actually used.
Practice a document-intelligence design exercise
Sketch a hypothetical document question-answering system with ingestion, retrieval, evaluation and deployment. Discuss trade-offs among relevant options named in the posting, such as Snowflake Cortex, Databricks Mosaic AI and OpenAI-compatible APIs; do not present the exercise as a company project.
Map the production lifecycle
Prepare a deployment diagram showing training, validation, model versioning, CI/CD and monitoring. Be ready to explain how you would investigate drift, performance, cost and latency while addressing security and responsible AI.
Prepare a cross-functional example
Outline a real instance of turning a business problem into an AI task with product or data-engineering partners. State the decisions made together and how the result was measured.
Write questions for the employer
Ask which use cases this role would tackle first, how Snowflake and Databricks fit the current architecture, and who owns model evaluation, deployment and production monitoring.

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…