New opportunity

Specialist Solutions Architect - AI/ML

Databricks · Mumbai, India

About this role

As an AI/ML Specialist Solutions Architect, you will be a technical expert for Databricks customers and the Field Engineering organization. Working with Solution Architects, you will help customers design production-grade ML and AI applications on the Databricks Data Intelligence Platform and align their technical roadmaps with the platform’s evolution. This is a hybrid role based in Mumbai, India.

Responsibilities

Architect, build, scale and optimize customer ML and AI workloads, including agents, end-to-end ML pipelines, training and inference optimization, cloud-native integrations and MLOps.
Advise enterprises on generative AI solutions such as RAG, tool-calling agents, multi-agent orchestration, guardrails, natural-language querying of structured data, AI evaluation, observability and monitoring.
Support Solution Architects during technical sales across feature engineering, training, tracking, serving and model monitoring. Share expertise with Databricks’ ML specialist community and mentor others.
Work with product and engineering teams to represent customer needs, influence priorities and the product roadmap, and support adoption of Databricks’ AI offerings.

Requirements

At least 5 years of hands-on industry ML experience in either of two areas: building and maintaining production-grade cloud infrastructure on AWS, Azure or GCP for deployed ML applications, including drift monitoring; or applying current LLM and agentic-system techniques, including vector databases, LLM fine-tuning, guardrails and deployment with tools such as Hugging Face, LangChain and OpenAI.
A graduate degree in a quantitative discipline such as computer science, engineering, statistics or operations research, or equivalent practical experience. Ability to communicate or teach technical concepts to both technical and non-technical audiences, collaborate, keep learning and drive business value through ML and AI.
Ability to meet technical training and role-specific expectations within 3 months of hire and travel up to 30% when needed.

Preferred

At least 2 years of customer-facing experience in a pre-sales or post-sales role.

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

Skills for this role

Machine LearningGenerative AIMLOpsLLMsRAGAI agentsMulti-agent orchestrationAI guardrailsAI evaluationAI observabilityModel monitoringDrift monitoringFeature engineeringModel trainingModel servingInference optimizationML pipelinesVector databasesLLM fine-tuningHugging FaceLangChainOpenAIAWSAzureGoogle Cloud PlatformDatabricksCloud infrastructureTechnical communicationCustomer-facing technical support

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

Production-grade customer architectures
Architect, build, scale and optimize ML and AI workloads on Databricks, including end-to-end pipelines, training and inference optimization, cloud-native integration and MLOps.
Enterprise generative AI design
Advise customers on RAG, tool-calling and multi-agent systems, guardrails, natural-language querying of structured data, evaluation, observability and monitoring.
Technical-sale support across the ML lifecycle
Support Solution Architects with advanced technical expertise from feature engineering and training through tracking, serving and model monitoring; communicate concepts to technical and non-technical audiences.
Customer advocacy and shared expertise
Represent customer needs to product and engineering teams to help shape priorities and adoption, while contributing to Databricks’ ML specialist community and mentoring others.

Likely challenges to think through

Balancing AI capability with production controls
Preparation hypothesis: An agent or RAG design may improve usefulness while adding evaluation, guardrail and monitoring complexity. Prepare to explain how you would compare designs against quality, latency and operational needs, then validate them before and after deployment.
Connecting architecture decisions across systems
Preparation hypothesis: A unified-platform design may still need cloud-native integrations and existing customer infrastructure. Prepare an example that explains integration boundaries, deployment and drift monitoring, and the trade-offs behind your choices without assuming a particular customer environment.
Maintaining depth while influencing multiple audiences
Preparation hypothesis: Supporting a technical sale while advising customers and relaying feedback to product teams requires different levels of detail. Prepare to explain one architecture decision to an engineer, a non-technical stakeholder and a product team, distinguishing verified customer needs from proposed improvements.

Practice your answers

Technical practice

How would you design an end-to-end ML pipeline for a production customer workload?

What to coverSuggested preparation (inferred): Explain how you would connect feature engineering, training, experiment tracking, serving and monitoring on a unified platform. Address reproducibility, deployment reliability and where you would place checks before a model reaches production.

How would you integrate a deployed ML application with cloud-native services?

What to coverSuggested preparation (inferred): Describe the infrastructure and operational boundaries you would consider on AWS, Azure or GCP. Discuss trade-offs involving scaling, access control and maintainability without assuming a particular customer architecture.

How would you detect and respond to drift in a production model?

What to coverSuggested preparation (inferred): Distinguish changes in input data from changes in model performance, and explain what you would monitor when labels arrive late. Describe how findings could trigger investigation, retraining or another response rather than an automatic change.

How would you optimize training and inference for an ML or AI workload?

What to coverSuggested preparation (inferred): Explain which measurements you would establish before changing a workload, such as training time, inference latency, throughput and cost. Discuss how an optimization might improve one measure while affecting model quality or reliability.

How would you design and evaluate a retrieval-augmented generation application?

What to coverSuggested preparation (inferred): Walk through document preparation, vector retrieval, context selection and answer generation. Explain how you would test retrieval quality separately from answer quality and handle responses without adequate supporting context.

When would you consider LLM fine-tuning instead of retrieval for a customer use case?

What to coverSuggested preparation (inferred): Compare what each approach changes and the evidence you would seek before choosing one. Address data requirements, evaluation, updating information and deployment complexity rather than presenting either approach as universally better.

How would you design a tool-calling agent, and when would multi-agent orchestration be warranted?

What to coverSuggested preparation (inferred): Explain tool selection, state handling, permissions and guardrails for an agent that can take actions. Compare a simpler single-agent design with multi-agent orchestration in terms of coordination, observability and failure modes.

How would you enable natural-language queries over structured enterprise data?

What to coverSuggested preparation (inferred): Discuss how a system could map user intent to the right data and produce a verifiable answer. Address ambiguity, access controls and checks against incorrect queries or misleading results.

How would you evaluate and monitor a production generative AI application?

What to coverSuggested preparation (inferred): Describe a mix of task-specific quality checks, safety checks and operational signals. Explain how traces or other observations would help diagnose whether a failure came from retrieval, an agent action or model output.

How would you support a Solution Architect in a technical discussion about model serving?

What to coverSuggested preparation (inferred): Show how you would clarify the customer's workload, success criteria and current constraints before proposing an architecture. Discuss serving and monitoring trade-offs in terms a technical buyer could use to evaluate a production design.

Behavioural practice

Tell me about a time you worked with another technical lead to resolve a customer's architecture concern.

What to coverSuggested practice (inferred): Use a real situation, explain your actions and collaboration, and state the result. Focus on how you clarified the concern and handled any technical trade-offs rather than claiming a particular sales outcome.

Describe a time you explained a complex ML or AI concept to people with different technical backgrounds.

What to coverSuggested practice (inferred): Give the situation, how you adjusted your explanation or teaching approach, and the result. Include how you checked understanding without overstating the audience's prior knowledge.

Tell me about a time you represented customer needs to a product or engineering team.

What to coverSuggested practice (inferred): Describe the situation, the evidence you gathered, the action you took and the result. Explain how you handled competing priorities without implying you controlled the product roadmap.

Describe a time you mentored a colleague or shared technical expertise across a team.

What to coverSuggested practice (inferred): Set out the situation, what you did to make the knowledge useful, and the result. Note how you adapted to the colleague's needs and what feedback showed whether the approach helped.

Tell me about a time you had to learn an unfamiliar AI or ML technique to address a practical need.

What to coverSuggested practice (inferred): Explain the situation, how you evaluated and applied the technique, and the result. Connect the learning to a measurable technical or business need, while acknowledging what remained uncertain.

Sharpen your knowledge

Topics to prepare

Production ML architecture
Compare designs for deploying an ML application on AWS, Azure or GCP, covering scaling, cloud-native integration and drift monitoring. The posting explicitly names production-grade cloud infrastructure as one of two qualifying experience paths.
RAG and agentic-system design
Sketch when to use RAG, a tool-calling agent or multi-agent orchestration; explain where vector databases and guardrails fit. These are explicit enterprise GenAI responsibilities, while the AI-engineering experience path includes vector databases, fine-tuning and guardrails.
The production ML lifecycle
Trace a model from feature engineering through training, tracking, serving and monitoring, and identify what can fail at each handoff. The role supports Solution Architects across these stages on a unified platform.
AI evaluation and observability
Draft an evaluation plan for a RAG application that distinguishes answer quality, guardrail behavior and operational monitoring. The posting explicitly calls for AI evaluation, observability and monitoring systems.
Customer roadmaps and platform trade-offs
Practice explaining how a customer's production goals could shape an ML and AI roadmap, then separate immediate architecture decisions from feedback for product and engineering teams. This reflects the stated customer-advisory and cross-functional responsibilities.

Skills to revise

Build an end-to-end ML pipeline
As a preparation exercise, implement a small pipeline covering feature preparation, training, tracking and serving, then explain its productionization gaps. The posting names end-to-end pipelines and technical-sale support across those stages; this exercise is a suggestion, not an employer requirement.
Prototype a guarded RAG workflow
As a preparation exercise, connect retrieval from a vector database to an LLM response, add a simple guardrail, and test a tool-calling path. The posting explicitly discusses RAG, vector databases, agents and guardrails; a prototype is suggested practice, not a stated deliverable.
Investigate model drift
As a preparation exercise, compare reference and recent input or prediction distributions, flag a meaningful change, and propose a monitoring response. Drift monitoring is explicit in the ML-engineering experience path.
Optimize a workload
As a preparation exercise, measure a training or inference bottleneck, change one design choice and report the effect and trade-off. Building, scaling and optimizing customer workloads—including training and inference optimization—are stated responsibilities.
Explain an architecture to two audiences
As a preparation exercise, present the same AI/ML design first to a Solution Architect with technical trade-offs, then to a non-technical stakeholder in terms of business value. The posting explicitly requires communication or teaching across both audiences and support during technical sales.

Your preparation checklist

Map your evidence to the required experience path
Prepare a one-page summary of 5+ years of hands-on industry ML work in at least one stated path: production cloud ML infrastructure with drift monitoring, or LLM and agentic systems. Note a quantitative graduate degree or equivalent practical experience; the posting does not require both experience paths.
Build a production workload walkthrough
Choose one real project you can discuss. Prepare a diagram and concise account of architecture, your decisions, deployment, scaling, monitoring, trade-offs and measurable outcomes. Where relevant, show feature engineering, training, tracking, serving or inference optimization.
Practice an enterprise GenAI design exercise
As interview preparation, sketch a RAG or tool-calling agent design with data flow, vector retrieval where appropriate, guardrails, evaluation, observability and monitoring. Explain design alternatives and what you would validate before production; these are topics named in the role, not a stated interview assignment.
Rehearse a customer-facing technical explanation
Prepare a short presentation of the walkthrough for both technical and non-technical audiences. Practice answering a Solution Architect’s questions about productionizing the workload and connecting technical choices to business value.
Prepare a customer-to-product example
Identify a real example you can discuss of gathering a customer need, collaborating across teams and turning feedback into a technical priority or roadmap recommendation. If you have pre- or post-sales experience, prepare evidence of it; the posting calls it preferred.
Bring focused questions about the role
Ask how the specialist partners with Solution Architects, what technical training and role-specific outcomes are expected in the first three months, and how customer feedback reaches product and engineering. Confirm practical expectations for the Mumbai hybrid role and travel of up to 30%.

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…