New opportunity

Exponentia.ai - GenAI Engineer - RAG/AIML

Exponentia.ai · Delhi, India

About this role

Exponentia.ai seeks a GenAI Engineer for its Delhi, India location. The role overview describes a GenAI Technical Lead who can build production-grade, enterprise-level generative AI solutions and lead proofs of concept (POCs) from scratch. The posting lists 3–8 years of experience.

Responsibilities

Design and develop end-to-end generative AI solutions, including retrieval-augmented generation (RAG) pipelines, LLM workflows and agent systems.
Lead POC development, implement prompt engineering strategies and build backend APIs with FastAPI or Flask.
Integrate AI solutions with enterprise platforms; deploy applications on Azure Web Apps, Function Apps or virtual machines; and use Databricks for pipelines and workflows.
Collaborate with stakeholders throughout solution development.

Requirements

Hands-on experience delivering production-level generative AI solutions, with RAG architecture, prompt engineering, LLM orchestration, Python backend development and end-to-end solution-building skills.
Experience with FastAPI or Flask, Azure OpenAI and application deployment, and Databricks pipelines and workflows. The posting also lists AI/ML, data modeling and deep learning among required skills.
Education/qualification: BTech, BSc or BCA.

Preferred skills include LangChain, LlamaIndex, vector databases, MLOps and CI/CD. Exponentia.ai describes opportunities to build its AI capability, collaborate across its international operations and access certifications, technology events and partner-led innovation labs.

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

Skills for this role

Generative AIRetrieval-Augmented Generation (RAG)Prompt EngineeringLarge Language Models (LLMs)LLM OrchestrationAI AgentsPythonFastAPIFlaskAzureAzure OpenAIAzure Web AppsAzure FunctionsAzure Virtual MachinesDatabricksData ModelingMachine LearningDeep LearningLangChainLlamaIndexVector DatabasesMLOpsCI/CDAPI Development

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

Own end-to-end GenAI delivery
The role overview calls for hands-on, production-grade enterprise solutions and leadership of proofs of concept from scratch. Prepare to explain how a solution moves from initial scope through implementation and deployment.
Design effective RAG pipelines
Building RAG pipelines and applying prompt engineering are explicit responsibilities. Prepare to explain retrieval and prompt-design choices, including how you would evaluate whether responses use the retrieved material well.
Build LLM workflows and agents
The posting names LLM workflows, agent systems and orchestration. Prepare to describe how you would structure multi-step behavior and decide when an agent is warranted rather than a simpler workflow.
Connect the application to enterprise systems
The role includes Python APIs using FastAPI or Flask, enterprise-platform integration, Azure deployment and Databricks pipelines. Prepare to trace how data, application logic and deployment fit together in one solution.
Lead POCs with stakeholders
Leading POCs from scratch and collaborating with stakeholders are stated duties. Prepare to describe how you would agree on a POC's scope and success criteria, then communicate what its results do and do not establish.

Likely challenges to think through

Balancing POC speed with production readiness
Preparation hypothesis: a quick demonstration may not establish that an enterprise solution is ready to deploy. Prepare an example of how you would choose what to validate early and what additional work would be needed before production.
Balancing answer quality with RAG complexity
Preparation hypothesis: adding retrieval and orchestration steps can improve grounding but make a workflow harder to diagnose and operate. Prepare to explain how you would compare a simple baseline with a more elaborate design.
Coordinating integration boundaries
Preparation hypothesis: connecting APIs, Azure-hosted applications, Databricks workflows and enterprise platforms may create handoff difficulties. Prepare to explain how you would define interfaces and test the full path, without assuming a particular employer architecture.

Practice your answers

Technical practice

How would you take an enterprise GenAI proof of concept from an initial idea to a production-ready solution?

What to coverSuggested preparation (inference): Walk through defining the use case, building an end-to-end prototype, evaluating it and deciding what must change before deployment. Explain trade-offs among answer quality, latency, cost and operational reliability; the posting calls for leading POCs and building production-grade solutions.

How would you design a RAG pipeline for enterprise content?

What to coverSuggested preparation (inference): Explain content preparation, data modeling, retrieval, context assembly and answer generation. Discuss how you would choose and update a retrieval index; vector databases are listed as preferred, not required.

How would you diagnose a RAG system that returns plausible but poorly supported answers?

What to coverSuggested preparation (inference): Separate retrieval failures from generation failures and describe tests that inspect retrieved context alongside the final answer. Explain what evidence would justify changing the data, retrieval approach or prompt.

How would you develop and assess prompts for an LLM workflow?

What to coverSuggested preparation (inference): Describe how you would specify the task, supply retrieved context and test prompts against representative inputs. Address the trade-off between more detailed instructions and their effects on consistency, latency or cost.

When would you use an agent system rather than a simpler LLM workflow?

What to coverSuggested preparation (inference): Compare a defined sequence of steps with an agent that can select actions, including the added complexity of controlling and evaluating those actions. The posting names both LLM workflows and agent systems; LangChain and LlamaIndex are preferred tools, not mandated choices.

How would you expose a GenAI solution through a Python API using FastAPI or Flask?

What to coverSuggested preparation (inference): Outline request validation, calling the LLM or RAG workflow, response handling and useful error reporting. Discuss how API design changes when model calls are slow or fail, since the role includes Python backend development.

How would you integrate an Azure OpenAI-based solution with an enterprise platform?

What to coverSuggested preparation (inference): Describe the boundaries between the enterprise system, backend API, model calls and data sources. Discuss practical concerns such as access control, data flow and handling integration failures without assuming a particular platform is used.

How would you choose among Azure Web Apps, Function Apps and virtual machines for a GenAI application?

What to coverSuggested preparation (inference): Compare deployment choices against the application's request pattern, runtime needs and operational overhead. Explain how you would verify that the deployed API and its model dependencies work reliably; the posting names these Azure options without prescribing one.

What role would Databricks pipelines and workflows play in an end-to-end GenAI solution?

What to coverSuggested preparation (inference): Describe how a data pipeline could prepare and refresh inputs consumed by a RAG application. Explain how you would detect stale or failed pipeline outputs and assess their effect on answers.

How would you manage changes to a production GenAI workflow?

What to coverSuggested preparation (inference): Explain how you would test and compare changes to prompts, retrieval and application code before release, then observe their effects afterward. MLOps and CI/CD are preferred skills in the posting, so treat specific release practices as design choices rather than stated requirements.

Behavioural practice

Tell me about a time you led a technical proof of concept from scratch.

What to coverSuggested preparation (inference): Use a real situation, the decisions and work you personally led, and the result. Focus on how you defined scope and judged whether the POC was ready to progress, reflecting the role's stated POC responsibility.

Describe a time stakeholder feedback changed the direction of a technical solution.

What to coverSuggested preparation (inference): Give the situation, how you clarified the underlying need and what you changed, then state the outcome. The posting calls for stakeholder collaboration, but does not specify a particular stakeholder process.

Tell me about a time you coordinated an integration across different systems or teams.

What to coverSuggested preparation (inference): Describe the situation, the dependency or interface you owned, the actions you took to align others and the result. Keep the example grounded in your actual experience rather than assuming the employer uses a particular enterprise platform.

Describe a time you had to balance delivery speed against technical quality.

What to coverSuggested preparation (inference): Explain the situation, the options you considered, your decision and its result. Connect the example to the posting's emphasis on both leading POCs and delivering production-grade solutions.

Tell me about a technical setback in a solution you owned and how you responded.

What to coverSuggested preparation (inference): State the situation and impact, the steps you took to diagnose or contain the issue, and the result. Choose a real example that shows how you communicated what you knew and adjusted your approach.

Sharpen your knowledge

Topics to prepare

RAG architecture
The posting requires RAG architecture. As revision, draw the path from source data through retrieval to an LLM response, and explain how you would diagnose an irrelevant retrieved passage.
Prompts, orchestration and agents
Prompt engineering and LLM orchestration are must-haves; LLM workflows and agent systems are responsibilities. As revision, compare when a single prompt, a multi-step workflow or an agent is appropriate, including how each could fail.
Production solution design
The role calls for production-grade, end-to-end enterprise GenAI solutions. As revision, outline how a POC would become a deployed service, including integration boundaries and how you would assess response quality and failures. Those assessment choices are preparation suggestions, not stated requirements.
Azure and Databricks architecture
The posting names Azure OpenAI, Azure Web Apps, Function Apps or VMs, and Databricks pipelines and workflows. As revision, sketch how data preparation, an API and an LLM call could fit across those named services.
AI/ML and data modeling foundations
AI/ML, data modeling and deep learning appear in the required skills. As revision, explain how data representation and model behavior could affect the quality of a GenAI solution; the posting does not specify a particular modeling technique.

Skills to revise

Build a small RAG pipeline
As a practice exercise, assemble a small document-to-answer flow and inspect whether retrieved passages support its answers. This rehearses the stated responsibility to build RAG pipelines; a vector database is preferred, not required.
Iterate on prompts and workflows
As a practice exercise, write two prompt variants for the same task, compare their outputs against a few test questions, then implement a simple multi-step LLM workflow. This addresses the explicit prompt-engineering and orchestration requirements.
Expose a Python API
As a practice exercise, build one endpoint with FastAPI or Flask that accepts a question and returns a structured response, including a handled failure case. Python backend expertise and FastAPI/Flask experience are explicit requirements.
Plan an Azure deployment
As a practice exercise, map an API that calls Azure OpenAI to one of the listed deployment options—Web Apps, Function Apps or VMs—and rehearse its deployment steps. Azure OpenAI and app deployment are named in the posting.
Outline a Databricks workflow
As a practice exercise, draft a Databricks pipeline that prepares source data for a GenAI application, identifying its inputs, outputs and a failure point. The role explicitly involves Databricks pipelines and workflows.
Present an end-to-end POC
As a practice exercise, prepare a short walkthrough covering a POC's objective, architecture, API, deployment path and questions for stakeholders. The posting calls for leading POCs from scratch and collaborating with stakeholders; this walkthrough is a preparation suggestion.

Your preparation checklist

Prepare a production GenAI walkthrough
The role calls for production-grade, end-to-end GenAI delivery. Prepare one project walkthrough covering the problem, your decisions and contributions, architecture, deployment, and observed results; distinguish a production system from a POC.
Sketch a RAG design
The posting specifically requires RAG architecture. Draw a sample pipeline from data ingestion through retrieval to response generation, and be ready to explain design trade-offs and how you would check answer quality.
Show prompt and orchestration decisions
Prepare a concrete example of prompt iteration and an LLM workflow or agent design. Explain why you chose that approach, how components interact, and what you would test.
Map the backend and deployment path
Prepare an architecture sketch connecting a Python FastAPI or Flask API to Azure OpenAI and an Azure deployment option named in the posting: Web Apps, Function Apps, or VMs. Explain your own implementation experience where applicable.
Explain a Databricks pipeline
The role names Databricks pipelines and workflows. Bring a diagram or concise example showing data inputs, processing steps, and how the pipeline supports an AI solution; label hypothetical elements clearly.
Rehearse leading a POC from scratch
The role overview emphasizes POC leadership and stakeholder collaboration. Outline how you would turn a stakeholder problem into a scoped demonstration, agree on success criteria, present findings, and decide what needs work before production.
Prepare targeted evidence and questions
List examples you can substantiate for the stated must-haves, plus any relevant preferred tools such as LangChain, LlamaIndex, vector databases, MLOps, or CI/CD. Ask how this role divides time between POCs and production delivery, and which Azure and Databricks components its projects use.

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…