New opportunity

Sr. AI/ML Engineer

ARMAKUNI INDIA · Ahmedabad, India

About this role

ARMAKUNI INDIA seeks a Sr. AI/ML Engineer in Ahmedabad to design, build, deploy, and maintain scalable AI/ML applications that address business challenges. The role spans generative AI, machine learning, data science, and computer vision.

Responsibilities

Architect generative AI and machine learning applications using AWS Cloud and other technologies. Work with LLM architecture, prompt engineering, custom LLM development and fine-tuning, LLM output evaluation, agentic workflows, and RAG pipelines integrated with traditional applications.
Solve data science problems through exploratory data analysis. Build automated data-cleaning, preprocessing, and model-retraining pipelines; track and validate experiments; and deploy, track, and monitor models with AWS SageMaker.
Address image classification, object detection, segmentation, and image captioning problems. Work with databases, data warehouses, and data lakes for storage and retrieval.

Qualifications and tools:

The posting specifies 3–7 years in NLP, data science, machine learning, and computer vision, plus 2–3 years implementing, deploying, and scaling generative AI solutions. Its summary describes 3–6 years of AI/ML experience.
Proficiency in Python; experience with LangChain, Langfuse, LlamaIndex, LangGraph, Crew AI, and LLM evaluation; and familiarity with AWS Bedrock, OpenAI GPT models, Claude, Llama, Gemini, and DeepSeek. Experience with vector databases such as Pinecone, OpenSearch, FAISS, and Chroma and their indexing mechanisms is required.
Knowledge of supervised and unsupervised learning, CNNs, RNNs, forecasting, time-series analysis, predictive analytics, classification, regression, and clustering. Tools listed include Scikit-learn, XGBoost, LightGBM, CatBoost, PyTorch, TensorFlow, Docker, Uvicorn, FastAPI, Flask, SQL, NoSQL, PostgreSQL, AWS DynamoDB, Redis, and Memcached.

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

Skills for this role

PythonGenerative AIMachine LearningDeep LearningData ScienceNatural Language ProcessingComputer VisionLarge Language ModelsPrompt EngineeringLLM Fine-TuningRAGAgentic WorkflowsLangChainLangGraphLangfuseCrew AILlamaIndexLLM EvaluationAWSAWS BedrockAWS SageMakerOpenAI GPT-4OpenAI GPT-4oOpenAI GPT-4o-miniClaudeLlamaGeminiDeepSeekPineconeOpenSearch FAISS?

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

Scalable generative AI applications
The posting calls for architecting generative AI and ML applications on AWS, building RAG pipelines, and integrating them with traditional applications. Prepare to explain architecture decisions from retrieval through application deployment.
LLM development and evaluation
The role includes prompt engineering, custom LLM development and fine-tuning, output evaluation, and agentic workflows. Be ready to explain how you would compare approaches and validate outputs rather than treating model choice as the result.
End-to-end ML operations
Responsibilities span exploratory data analysis, automated cleaning and preprocessing, retraining, experiment tracking and validation, and deployment and monitoring with SageMaker. Focus preparation on how these stages connect in a maintainable pipeline.
Breadth across ML and computer vision
The posting covers classification, forecasting and other ML methods alongside image classification, detection, segmentation, and captioning. It also calls for efficient storage and retrieval using databases, warehouses, and data lakes. Prepare to explain how the problem and data would guide the method.

Likely challenges to think through

Balancing retrieval quality and efficiency
Preparation hypothesis, not a stated employer problem: a RAG application may need to balance retrieval relevance against indexing complexity and response time. Prepare to discuss how you would choose an indexing approach, evaluate retrieved results, and decide whether changes improve the application.
Knowing when to customize an LLM
Preparation hypothesis: prompts, RAG, fine-tuning, and agentic workflows offer different ways to improve an application, with differing complexity and evaluation needs. Prepare a decision framework tied to the task and measurable output quality, rather than assuming the most elaborate approach is best.
Keeping automated models trustworthy
Preparation hypothesis: automated retraining can speed updates but risks promoting a model without adequate validation. Prepare to explain experiment comparisons, deployment checks, and monitoring signals in the SageMaker-based lifecycle described in the posting.

Practice your answers

Technical practice

How would you decide between prompting, fine-tuning, and building a custom LLM for an application?

What to coverThe role covers prompt engineering, fine-tuning, custom LLMs, and LLM internals. Preparation suggestion: Explain what each approach changes, how you would evaluate the resulting outputs, and the trade-offs in data needs and operational effort.

How would you design a RAG pipeline that integrates with an existing application?

What to coverThe posting calls for RAG, vector databases, indexing mechanisms, and integration with traditional applications. Preparation suggestion: Walk through document preparation, indexing, retrieval, and response generation, including how you would test retrieval quality and handle stale information.

When would you use an agentic workflow rather than a simpler LLM pipeline?

What to coverAgentic workflows and tools including LangGraph and Crew AI appear in the posting. Preparation suggestion: Describe a suitable workflow, how you would constrain and observe its steps, and when added complexity would not be justified.

How would you evaluate and troubleshoot the outputs of a generative AI application?

What to coverThe role explicitly includes LLM output evaluation and lists Langfuse. Preparation suggestion: Discuss test cases, criteria for output quality, and how recorded results could help distinguish retrieval, prompt, and model failures.

How would you architect and scale a generative AI application on AWS?

What to coverThe posting calls for scalable AWS applications and mentions Bedrock alongside several LLM options. Preparation suggestion: Explain how you would select a model and deployment approach, then assess latency, reliability, and cost as usage grows.

How would you build an automated pipeline from exploratory analysis to model retraining?

What to coverThe role includes EDA, data cleaning, preprocessing, and automated retraining. Preparation suggestion: Describe how you would find data-quality issues, prevent leakage, and validate both data and model changes before deployment.

How would you track experiments and monitor a deployed model in SageMaker?

What to coverThe posting specifies experiment tracking, validation, and SageMaker training, deployment, and monitoring. Preparation suggestion: Explain which inputs, model versions, and results you would record, and how monitoring findings would inform investigation or retraining.

How would you choose and validate a model for classification, clustering, or time-series forecasting?

What to coverThe posting spans supervised and unsupervised learning, predictive analytics, and time-series analysis. Preparation suggestion: Compare appropriate validation approaches and metrics for these different tasks, including why temporal ordering matters for forecasting.

How would your approach differ across image classification, detection, segmentation, and captioning?

What to coverAll four computer-vision tasks are named in the role, along with CNNs and deep-learning frameworks. Preparation suggestion: Contrast their outputs, annotation needs, and evaluation methods, and explain how error analysis would guide improvements.

How would you serve an ML model through an API while managing storage and retrieval?

What to coverThe posting lists REST API tools, Docker, SQL and NoSQL databases, data lakes and warehouses, and caching technologies. Preparation suggestion: Sketch a Python-based serving path and explain where persistent storage or caching would help, including the trade-off between response speed and fresh data.

Behavioural practice

Tell me about a time you turned a business challenge into an AI/ML problem.

What to coverThe role is framed around AI/ML applications that solve business challenges. Preparation suggestion: Use a real situation, your actions in defining the problem and success measure, and the result; distinguish what the model achieved from its business effect.

Tell me about a time you integrated an AI capability with an existing application.

What to coverThe posting specifically calls for RAG pipelines integrated with traditional applications. Preparation suggestion: Give a real situation, the actions you took to coordinate integration and resolve constraints, and the resulting application behaviour.

Tell me about a time you investigated a deployed model or application that was underperforming.

What to coverDeploying, tracking, monitoring, and maintaining AI/ML applications are stated responsibilities. Preparation suggestion: Describe the observed problem, your diagnostic actions, and the result, including what you could and could not conclude from the evidence.

Tell me about a time experiment results changed your approach.

What to coverThe role includes EDA, experiment tracking, and model validation. Preparation suggestion: Use a real situation to explain what the evidence showed, what you changed, and the result rather than presenting an unverified improvement.

Tell me about a time you balanced model quality against deployment constraints.

What to coverThe posting emphasizes scalable, efficient applications as well as model evaluation and deployment. Preparation suggestion: Describe a real situation, the options and trade-offs you considered, your action, and the observed result.

Sharpen your knowledge

Topics to prepare

LLM architecture and adaptation
The posting calls for knowledge of LLM internals, prompt engineering, custom LLMs, and fine-tuning. As preparation, sketch how an LLM produces a response and compare when prompting versus fine-tuning would address a business task.
RAG and vector indexing
The role explicitly includes RAG pipelines, vector databases, and indexing mechanisms. As preparation, diagram document ingestion, indexing, retrieval, and answer generation; explain how indexing choices affect retrieval quality and latency.
Agentic workflows and evaluation
The posting names agentic workflows and LLM output evaluation. As preparation, outline a multi-step workflow and define checks for answer quality, tool-use errors, and failure handling.
ML fundamentals and forecasting
The posting specifies supervised and unsupervised learning, classification, regression, clustering, forecasting, and time-series analysis. As preparation, choose a model and validation approach for each of a classification and forecasting problem, including how to avoid time leakage.
Computer vision approaches
The role covers image classification, object detection, segmentation, and captioning, and names CNNs. As preparation, compare the output and evaluation needs of those four tasks and explain when a CNN-based approach fits.
Scalable ML lifecycle
The posting emphasizes scalable applications, automated retraining, experiment validation, deployment, and monitoring. As preparation, draw a lifecycle from data ingestion through monitored deployment and identify what should trigger investigation or retraining.

Skills to revise

Prototype prompting and model adaptation
As a practice exercise, use Python to test prompt variants against a small fixed set of examples, record outputs, and outline a fine-tuning experiment with a held-out evaluation set. This rehearses the posting’s prompt-engineering, custom-LLM, fine-tuning, and evaluation work.
Build a small RAG pipeline
As a practice exercise, index a small document set with FAISS or Chroma, retrieve passages for test questions, and connect the results to an LLM response. Inspect failed retrievals. This directly reflects the requested RAG and vector-database experience.
Trace an agentic workflow
As a practice exercise, implement a bounded workflow with LangGraph or Crew AI and inspect traces with Langfuse; test an unsuccessful step and a recovery path. The posting names these tools, agentic workflows, and LLM output evaluations.
Run a reproducible ML experiment
As a practice exercise, use Python and Scikit-learn or XGBoost to perform EDA, clean data, establish a baseline, and validate a model; then outline how training, deployment, and monitoring would run in SageMaker. This targets the stated data-science and ML lifecycle responsibilities.
Implement a vision baseline
As a practice exercise, train or adapt a small image classifier in PyTorch or TensorFlow, inspect error cases, and explain what would change for detection or segmentation. This targets the posting’s computer-vision tasks and deep-learning frameworks.
Serve an application endpoint
As a practice exercise, expose a model or RAG pipeline through FastAPI or Flask, package it with Docker, and sketch where PostgreSQL or DynamoDB and Redis or Memcached would fit. This reflects the listed API, container, database, and caching technologies; the particular combination is a preparation suggestion, not a specified employer stack.

Your preparation checklist

Map the role to evidence
Create a one-page map of the posting’s generative AI, ML, computer vision, Python, and AWS requirements. Note which examples you can discuss and which topics need review; do not imply experience you cannot substantiate.
Prepare a RAG project walkthrough
If you have a relevant project, outline its business problem, architecture, retrieval and indexing choices, application integration, deployment, and results. Otherwise, prepare a clearly labeled design example. The posting specifically calls for scalable RAG applications.
Practice explaining LLM decisions
Prepare a short comparison of prompting, fine-tuning, output evaluation, and agentic workflows. Show how you would test answer quality and choose among the LLMs and frameworks named in the posting without claiming to have used every one.
Work through an ML lifecycle example
Prepare a walkthrough or design covering EDA, automated cleaning and preprocessing, experiment tracking, validation, retraining, and SageMaker deployment and monitoring. Identify the metrics you would use and why.
Review one computer vision problem
Choose image classification, object detection, segmentation, or image captioning from the posting. Prepare an example or design explaining data preparation, model choice, evaluation, and likely failure cases.
Sketch a production architecture
As a preparation exercise, diagram how a Python API, AWS services, model serving, and storage or retrieval could fit together. Be ready to explain scaling and monitoring trade-offs; the posting names these technologies but does not prescribe one architecture.
Draft questions for the employer
Ask which business problems are the immediate priority, how generative AI and conventional ML work are divided, what evaluation and monitoring are in place, and how the stated experience ranges in the summary and qualifications should be interpreted.

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…