New opportunity

Senior Solutions Architect, Generative AI

NVIDIA · Gurugram, India

About this role

NVIDIA seeks a Senior Solutions Architect, Generative AI in Gurugram, India, to design and deliver customer solutions centered on large language models (LLMs), agentic AI and retrieval-augmented generation (RAG). The role combines solution architecture, hands-on LLM training and optimization, customer engagement, and technical leadership.

Responsibilities

Architect end-to-end generative AI solutions tailored to customers’ language-related business challenges. Design and implement agentic and RAG workflows for content generation and information retrieval, and help customers integrate them into their applications and systems.
Lead customer workshops and design sessions. Train and optimize LLMs using NVIDIA hardware and software platforms, developing efficient training strategies and guiding best practices for LLM training and RAG implementation.
Support pre-sales work with sales and business development teams through technical presentations and demonstrations. Share customer feedback with NVIDIA engineering teams and stay current with developments in language models and generative AI.

Required qualifications

B.Tech, master’s or Ph.D. in Computer Science or Artificial Intelligence, or equivalent experience; 10+ years of hands-on experience in a technical role focused on generative AI, particularly LLM training.
Demonstrated success deploying and optimizing LLMs for production inference; understanding of models such as GPT-3 and BERT; and expertise in training and fine-tuning with TensorFlow, PyTorch or Hugging Face Transformers.
Proficiency in deployment and inference optimization across hardware platforms, especially GPUs; knowledge of GPU cluster architecture and parallel processing; and strong communication, collaboration, workshop leadership and technical presentation skills.

Preferred qualifications

Ability to improve inference speed, memory efficiency and resource utilization; familiarity with Docker and Kubernetes; deeper knowledge of parallel and distributed computing; and hands-on experience with NVIDIA GPUs, GPU cluster management, and scalable LLM training and inference workflows.

The posting mentions competitive salaries and a benefits package but provides no compensation figures.

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 ModelsAgentic AIRetrieval-Augmented GenerationLLM trainingLLM fine-tuningLLM inferenceModel deploymentModel optimizationGPT-3BERTTensorFlowPyTorchHugging Face TransformersGPU cluster architectureParallel processingCustomer engagementTechnical presentationsWorkshop facilitationDockerKubernetesDistributed computingGPU cluster management

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

Customer-led solution architecture
The role calls for end-to-end generative AI solutions tailored to customers’ language-related challenges. Prepare to explain how you would turn a customer workshop into an architecture and integration plan.
Agentic and RAG workflows
NVIDIA specifies designing and implementing agentic and RAG workflows, including RAG for content generation and information retrieval, and integrating them into customer applications. Be ready to explain workflow choices and integration boundaries.
LLM training and optimization
The posting emphasizes leading LLM training and optimization on NVIDIA hardware and software, with expertise in training and fine-tuning frameworks. Prepare to explain an efficient training strategy and the decisions behind it.
Production inference on GPUs
Production LLM deployment and inference optimization are required, alongside GPU-cluster and parallel-processing knowledge. Prepare to discuss how you would measure performance and improve inference efficiency; speed, memory and resource utilization are specifically highlighted as preferred strengths.
Technical leadership across teams
The role includes leading design sessions, supporting pre-sales demonstrations, guiding customers on best practices and conveying feedback to NVIDIA engineering. Prepare to communicate the same technical approach clearly to both technical and non-technical audiences.

Likely challenges to think through

Choosing workflow complexity
Preparation hypothesis, not a stated NVIDIA problem: a customer may need information retrieval without the added complexity of an agentic workflow. Prepare criteria for choosing an appropriate RAG or agentic design and explaining its integration implications.
Balancing inference constraints
Preparation hypothesis: improving inference speed may compete with memory efficiency and resource utilization, especially at GPU-cluster scale. Prepare an example of how you would establish a baseline, compare options and justify the resulting trade-off.
Aligning demonstrations with delivery
Preparation hypothesis: a compelling pre-sales demonstration may not establish that a solution is ready for production integration. Prepare to distinguish demonstrated capabilities from deployment assumptions, and to show how customer feedback would inform subsequent design and engineering discussions.

Practice your answers

Technical practice

How would you architect an end-to-end generative AI solution for a customer's language-related challenge?

What to coverSuggested preparation: Explain how you would clarify the business objective, application constraints and success criteria before choosing an LLM, RAG or agentic workflow. Describe the components and trade-offs you would discuss with the customer.

When would you choose a GPT-3-like model rather than a BERT-like model?

What to coverSuggested preparation: Compare their typical strengths for generation and language understanding, and explain how the customer's task would shape the choice. Address what you would validate before committing to an architecture.

How would you plan and optimize LLM training or fine-tuning?

What to coverSuggested preparation: Walk through the training objective, data preparation, validation and iteration using a framework such as PyTorch, TensorFlow or Hugging Face Transformers. Explain how you would balance model quality against compute and memory costs.

How would you scale LLM training across a GPU cluster?

What to coverSuggested preparation: Explain relevant parallel and distributed computing approaches, including where communication or memory can limit scaling. Describe how you would evaluate whether additional GPUs improve training efficiency.

How would you design a RAG workflow for information retrieval and content generation?

What to coverSuggested preparation: Describe how relevant content reaches the model and how you would assess both retrieval quality and the resulting answers. Discuss trade-offs between relevance, response quality and latency.

How would you decide whether an agentic workflow is appropriate?

What to coverSuggested preparation: Explain what an agentic design would add to a customer's language-related task compared with a simpler LLM or RAG workflow. Discuss how you would evaluate its reliability and complexity before recommending it.

How would you integrate a RAG workflow into an existing customer application?

What to coverSuggested preparation: Describe the application interfaces, content access and operational constraints you would establish with the customer. Explain how you would validate that the integrated workflow meets the original use case.

How would you deploy an LLM for production inference on GPUs?

What to coverSuggested preparation: Outline the path from a validated model to a serving configuration, including how you would measure latency, throughput, memory use and resource utilization. Explain the trade-offs you would make when optimizing those measures.

How would you troubleshoot slow or memory-constrained LLM inference?

What to coverSuggested preparation: Explain how you would distinguish model, serving and GPU-resource bottlenecks using measurements rather than assumptions. Describe how you would test whether a change improves performance without undermining output quality.

What role could Docker and Kubernetes play in scalable model deployment?

What to coverSuggested preparation: These tools appear as preferred familiarity, not required expertise; explain their possible roles in packaging and orchestrating inference services. Discuss when their operational complexity would be justified.

Behavioural practice

Tell me about a time you translated an unclear customer need into a technical design.

What to coverSuggested preparation: Use a real situation, the questions you asked and the design decisions you made. Close with the result and how you established whether the solution addressed the customer's need.

Describe a workshop you led with stakeholders who had different technical backgrounds.

What to coverSuggested preparation: Set out the situation, how you structured the session and adapted your explanations, and what decisions or next steps resulted. Focus on your own actions rather than claiming consensus without evidence.

Give an example of supporting a technical presentation or demonstration alongside a commercial team.

What to coverSuggested preparation: Explain the audience and objective, your contribution to presenting technical capabilities, and the outcome. Include how you handled questions or limitations accurately.

Tell me about a time you relayed customer feedback to an engineering team.

What to coverSuggested preparation: Describe the customer issue, how you made the feedback actionable and how you collaborated on a response. State the observed result, including any unresolved constraints.

Describe a time you guided others through a difficult model-training or RAG design trade-off.

What to coverSuggested preparation: Explain the situation, the options and evidence you used, and how you helped the group reach a decision. Finish with the result and what you learned from it.

Sharpen your knowledge

Topics to prepare

Language-model architectures
Compare GPT-3, BERT and similar architectures: explain how their differing approaches affect a language-generation or information-retrieval solution. The posting explicitly calls for in-depth understanding of these models.
LLM training and fine-tuning strategy
For a hypothetical customer language task, outline when to train, fine-tune or use an existing model, and explain how you would judge training effectiveness. Leading efficient LLM training is central to the role.
Agentic and RAG solution design
Draw an end-to-end workflow showing retrieval, generation, agent decision points and integration with a customer application; identify where information quality could fail. The role explicitly involves architecting and implementing these workflows.
GPU clusters and parallel processing
Explain, with a cluster diagram, how parallel processing can accelerate LLM training and inference and where resource constraints arise. GPU cluster architecture is a required knowledge area; deeper distributed-computing knowledge is preferred.
Production inference trade-offs
Work through a hypothetical deployment choice across hardware platforms, explaining effects on inference speed, memory efficiency and resource use. Production LLM deployment and GPU-focused optimization are required; improvement on those three measures is highlighted as preferred.

Skills to revise

Fine-tune an LLM
Using one framework named in the posting—TensorFlow, PyTorch or Hugging Face Transformers—rehearse a small fine-tuning run and record the training choices and results you would explain to a customer. Hands-on training and fine-tuning expertise is required.
Build a RAG workflow
Implement a small retrieval-to-generation example using sample content, then test whether responses use the retrieved information and describe how the workflow would connect to an application. The posting explicitly calls for RAG implementation and customer integration.
Measure and optimize inference
Run a representative model inference workload, record speed and memory use, make one optimization change and compare the results. This rehearses the required production-inference optimization work without assuming a particular deployment tool.
Plan a GPU-cluster workflow
If GPU-cluster access is available, test a small parallel training or inference workload and inspect resource utilization; otherwise, specify the execution plan and measurements you would use. Leveraging parallel processing is required, while hands-on NVIDIA GPU cluster management is preferred.
Lead a customer design session
Rehearse a workshop: elicit a language-related business challenge, sketch an LLM or RAG solution, and deliver a short technical demonstration explanation for both technical and non-technical listeners. Customer workshops, presentations and pre-sales demonstrations are stated responsibilities.

Your preparation checklist

Map evidence to the role
Make a one-page inventory of relevant work in LLM training, production inference, GPU architecture and customer-facing leadership. The posting asks for 10+ years of technical experience and a track record in these areas; use only work you can substantiate.
Build an end-to-end solution walkthrough
Prepare a diagram and brief explanation of a real project, if available, showing the customer’s language-related problem, LLM and RAG or agentic workflow, application integration, design decisions and outcome. These reflect the role’s stated architecture responsibilities.
Explain a training and fine-tuning decision
Choose a real LLM training example and document the model choice, TensorFlow, PyTorch or Hugging Face workflow used, efficiency decisions and observed results. Be ready to explain how the approach would use NVIDIA hardware and software without implying experience you do not have.
Show production inference trade-offs
Prepare a before-and-after account of a deployment or optimization you worked on, with available measurements for speed, memory use and resource utilization. Explain GPU-related constraints and what changed; production inference optimization is a stated requirement.
Practice a GPU-scale design exercise
Sketch how you would scale LLM training and inference across a GPU cluster, including parallel-processing choices and deployment trade-offs. If relevant to your experience, discuss NVIDIA GPUs, Docker or Kubernetes; the posting identifies the latter tools as preferred, not required.
Rehearse a customer design session
Draft discovery questions and a short, audience-appropriate explanation of an LLM and RAG solution, then practice presenting it to both technical and non-technical listeners. The role includes workshops, demonstrations and pre-sales collaboration.
Prepare questions for NVIDIA
Bring questions about the customer problems this role tackles, how solution architects work with sales and engineering, and which NVIDIA platforms and GPU environments the team uses. These are questions to clarify the role, not assumptions about its processes.

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…