New opportunity

Lead Architect – Full-Stack, Cloud, Data & AI Engineering

Fractal Analytics Ltd. · Mumbai, India

About this role

The Lead Architect owns technical direction for enterprise agentic AI solutions across application, cloud, data and AI layers. The role leads delivery through the team, establishing standards and reusable deployment assets so Forward Deployed Engineers (FDEs) can build, deploy and operate solutions in client environments independently. Hands-on work is selective, focused on prototyping high-risk components and removing critical blockers rather than sustained feature delivery. The posting lists Mumbai, Bengaluru, Pune, Chennai and Gurgaon as locations.

Responsibilities

Set target architecture and decide where agents add value versus deterministic rules or workflows. Guide full-stack patterns for Python services, JavaScript/TypeScript front ends, SQL and NoSQL data design, APIs, CI/CD and DevOps.
Own Azure architecture decisions covering service selection, identity, networking, environments, non-functional targets and cloud cost. Direct and review PySpark and Databricks pipeline architecture, layered data design, quality controls and performance standards.
Own agent and orchestration design, evaluation harnesses, guardrails, human-approval flows, tracing, versioning and drift monitoring.
Break work into buildable increments; lead design walkthroughs and code reviews; and set coding, testing, release and documentation standards. Create reference architecture, infrastructure-as-code modules, pipeline templates, runbooks and environment checklists. Pair with and mentor FDEs, transferring deployment ownership against defined competency milestones.
Defend architecture and security posture in client reviews, act as final technical escalation for deployment and production issues, and review production code as needed.

Required experience: At least 10 years in software, platform or applied AI engineering, including at least 4 years leading engineering teams whose systems reached production. Requires full-stack delivery experience; hands-on architecture experience with client cloud and security teams; working depth in PySpark, Databricks and agent development using a mainstream orchestration framework, evaluation and production monitoring; and a demonstrated record of mentoring engineers.

Success is measured by independent FDE deployments, reduced time to deploy through reusable assets, timely production delivery with minimal architecture or security remediation, and tracking agent quality, availability, latency and cloud cost against baselines while catching regressions before release.

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

Skills for this role

AI agentsAgent orchestrationAIOpsPythonJavaScriptTypeScriptSQLNoSQLAPIsCI/CDDevOpsAzurePySparkDatabricksData engineeringCloud architectureInfrastructure as CodeEvaluation harnessesAI guardrailsHuman-approval flowsTracingDrift monitoringProduction monitoringCode reviewTechnical mentoringSecurity architecture

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

Set the end-to-end technical direction
Own the target architecture across application, cloud, data and AI layers, including decisions about where agents add value and where deterministic rules or workflows suffice.
Own Azure and security decisions
Decide service selection, identity, networking, environments, non-functional targets and cloud cost; defend the architecture and security posture in client reviews.
Guide full-stack and data engineering
Set patterns for Python services, JavaScript/TypeScript front ends, data stores, APIs and CI/CD. Direct and review PySpark and Databricks pipeline design, quality controls and performance standards.
Make agent behavior measurable and governable
Own orchestration design, evaluation harnesses, guardrails, human-approval flows, tracing, versioning and drift monitoring. Success measures include tracking quality, availability, latency and cost against baselines and catching regressions before release.
Build independent deployment capability
Create reusable architecture, infrastructure-as-code modules, pipeline templates, runbooks and checklists. Mentor Forward Deployed Engineers against competency milestones so they can deploy and operate without escalation.

Likely challenges to think through

Choosing agents selectively
Likely preparation challenge, not a confirmed employer problem: agent flexibility may not justify its evaluation, latency and operating costs when a rule or workflow would suffice. Prepare a decision framework that compares the options against reliability, oversight and measurable value.
Reusing assets across client environments
Likely preparation challenge: standard templates can shorten deployment, while client identity, networking and security constraints may require changes. Prepare an example of what should remain standardized, what should be configurable and how to validate the result in a client review.
Staying hands-on without becoming the bottleneck
Likely preparation challenge: the architect is the final technical escalation but is expected to transfer deployment ownership to FDEs. Prepare an approach to risk-based prototyping, review, pairing and milestone-based handover that preserves team autonomy.
Balancing production measures
Likely preparation challenge: improving agent quality can affect latency, availability or cloud cost. Prepare a way to set baselines, evaluate changes before release and decide when a regression warrants blocking deployment.

Practice your answers

Technical practice

When would you use an agent rather than a deterministic workflow?

What to coverSuggested preparation: Explain how you would judge whether an agent adds value over rules or a fixed workflow. Address predictability, operational risk, evaluation effort and the point at which you would choose the simpler approach.

How would you set the target Azure architecture for an enterprise agentic AI solution?

What to coverSuggested preparation: Walk through service-selection decisions across application, data and AI layers, including environments and non-functional targets. Explain how you would weigh availability, latency and cloud cost rather than naming services without a rationale.

How would you defend cloud identity, networking and security decisions in a client review?

What to coverSuggested preparation: Describe the architecture decisions and evidence you would bring to a review with client cloud and security teams. Explain how you would handle a requested change while protecting the agreed security posture and delivery timeline.

What full-stack patterns would you standardize for the team?

What to coverSuggested preparation: Cover boundaries between Python services, JavaScript or TypeScript front ends, APIs, and relational or NoSQL stores. Explain which decisions merit shared standards and where teams need flexibility, including implications for testing and deployment.

How would you review a PySpark and Databricks pipeline architecture?

What to coverSuggested preparation: Discuss layered data design, quality controls and performance standards, including how you would diagnose a pipeline that becomes slower as data grows. Distinguish architectural issues from implementation issues you would ask the team to investigate.

How would you design an agent orchestration flow with human oversight?

What to coverSuggested preparation: Explain how you would define orchestration boundaries, guardrails and human-approval points. Address how tracing and versioning would help operators understand decisions and investigate failures.

How would you evaluate agent changes before release and monitor them afterward?

What to coverSuggested preparation: Describe an evaluation harness, defined quality baselines and checks for regressions before release. Explain how you would track quality, availability, latency and drift in production, and decide when a change needs investigation.

How would you turn a person-dependent deployment into a repeatable practice?

What to coverSuggested preparation: Explain the roles of a reference architecture, infrastructure-as-code modules, pipeline templates, runbooks and environment checklists. Discuss how you would test whether these assets reduce time to deploy and let FDEs deploy without escalation.

How would you set engineering standards without becoming the delivery bottleneck?

What to coverSuggested preparation: Describe how you would break work into buildable increments and use design walkthroughs and code reviews to enforce coding, testing, release and documentation standards. Explain which high-risk component you would prototype yourself versus leave to the team.

How would you handle a critical production deployment issue?

What to coverSuggested preparation: Outline how you would establish the failure's scope, use available tracing and operational evidence, and prioritize restoring service. Explain how you would involve the team, review production code when necessary and turn the resolution into reusable guidance.

Behavioural practice

Tell me about mentoring engineers until they could deploy independently.

What to coverSuggested preparation: Give a real situation, the pairing, design-review or enablement actions you took, and the result. Explain what evidence showed they could deploy and operate without relying on you.

Describe a time you defended an architecture decision with security or cloud stakeholders.

What to coverSuggested preparation: Set out the situation and the stakeholders' concern, then explain your rationale and how you responded to objections. Close with the decision and its effect on remediation or delivery.

Tell me about leading a team from architecture through production delivery.

What to coverSuggested preparation: Describe the situation, how you divided the work and used reviews to maintain standards, and the production result. Be clear about what the team delivered versus what you personally implemented.

Describe a time you stepped in to remove a critical technical blocker.

What to coverSuggested preparation: Explain the situation, why selective hands-on work was warranted and what you did to unblock the team. State the result and how you returned ownership to others rather than becoming the ongoing implementer.

Tell me about making a deployment process less dependent on one person.

What to coverSuggested preparation: Describe the initial dependency, the reusable assets or documentation you introduced, and how others adopted them. Give the result using evidence such as independent deployments or reduced time to deploy, if available from your real example.

Sharpen your knowledge

Topics to prepare

Agentic versus deterministic architecture
The role owns decisions about where agents add value versus rules or workflows. Suggested revision: take a hypothetical enterprise workflow, classify its steps as agentic or deterministic, and explain the trade-offs and approval points.
Azure architecture and security posture
The posting assigns ownership of Azure service selection, identity, networking, environments, non-functional targets and cost. Suggested revision: sketch a target architecture and rehearse how you would defend its security and cost decisions in a client review.
Data pipeline design
The role directs PySpark and Databricks architecture, including layered data design, quality controls and performance standards. Suggested revision: outline a layered pipeline and explain where you would check quality and diagnose poor performance.
Agent evaluation and operations
The posting calls for evaluation harnesses, guardrails, human approval, tracing, versioning and drift monitoring. Suggested revision: define release checks and production baselines for agent quality, availability, latency and cloud cost, then explain how a regression would be caught before release.
Full-stack delivery standards
The architect guides patterns for Python services, JavaScript/TypeScript front ends, SQL and NoSQL stores, APIs, CI/CD and DevOps. Suggested revision: trace a change across those layers and identify the design, testing and release decisions an architect should review.
Leading FDEs to independent deployment
The role leads through design reviews, reusable assets and mentoring rather than sustained feature delivery. Suggested revision: map a hypothetical deployment into buildable increments and define competency milestones for handing build, deployment and operational ownership to FDEs.

Skills to revise

Draft a reviewable reference architecture
Suggested exercise: draw the application, Azure, data and agent layers; annotate identity, networking, environments, APIs and non-functional targets. This practices the target-architecture ownership and client-review responsibilities stated in the posting.
Prototype and evaluate a high-risk agent flow
Suggested exercise: use a mainstream orchestration framework to prototype one narrow agent flow, with a deterministic alternative, an approval step, trace records and a small evaluation set. The posting expects working depth in agent development and selective prototyping to unblock the team.
Review a PySpark pipeline
Suggested exercise: inspect or write a small PySpark transformation for a layered Databricks pipeline, then document data-quality checks and a performance-review plan. The role requires working depth in both PySpark and Databricks while directing and reviewing pipeline work.
Conduct a cross-layer code and release review
Suggested exercise: review a hypothetical Python API and JavaScript/TypeScript front-end change alongside its SQL or NoSQL design and CI/CD release steps; record specific review findings. The posting makes code reviews and full-stack standards part of the architect’s remit.
Build a reusable deployment handover
Suggested exercise: draft an environment checklist, runbook and pipeline-template outline, then walk through a deployment handover against competency milestones. These are named deployment assets, and independent FDE deployment is an explicit success measure.

Your preparation checklist

Prepare a production-delivery walkthrough
Select a relevant example, if available, and document your architecture decisions, team leadership, client reviews, production outcome and your selective hands-on contribution. The posting requires 10+ years in engineering, including 4+ years leading teams whose systems reached production; prepare a factual timeline rather than assuming your experience meets it.
Sketch an end-to-end architecture
As a preparation exercise, draw a client-environment solution spanning Python services, a JavaScript/TypeScript front end, APIs, SQL and NoSQL stores, CI/CD and deployment. Explain which decisions you would own and which work you would guide or review.
Defend an Azure design
Prepare a short decision record covering service selection, identity, networking, environments, non-functional targets, security-review trade-offs and cloud cost. Rehearse how you would explain it to client cloud and security teams.
Walk through a data-pipeline design
Sketch a PySpark and Databricks pipeline with layered data design, quality controls and performance standards. Identify what you would direct, what you would review and how you would check production readiness.
Make the agent-versus-workflow case
Prepare a design exercise that explains where an agent adds value and where deterministic rules suffice. Include an orchestration approach, evaluation harness, guardrails, human approval, tracing, versioning, drift monitoring and pre-release regression checks.
Show how you would make deployments repeatable
Outline a reference architecture, IaC modules, pipeline templates, runbooks and environment checklist. Pair these with an example enablement plan showing how FDEs could take deployment and operating ownership against competency milestones.
Draft outcome-focused questions
Ask how the team currently measures independent FDE deployments, time-to-deploy, architecture and security remediation, and baselines for agent quality, availability, latency and cloud cost. These are stated success measures; the current baselines are not given.

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…