New opportunity

Machine Learning Engineer

Coinbase · Remote - India

About this role

Coinbase’s CX Intelligence team is hiring an IC4 Machine Learning Engineer to develop a hybrid vendor-internal chatbot experience for customer support. The team operates a multi-agent platform connecting the Help Center, Coinbase chat, and agent workflows. The role focuses on a unified orchestration layer that coordinates vendor AI, internal multi-agent systems, and human participants so customers can get answers faster and support agents can resolve cases more effectively.

Responsibilities

Build and improve orchestration for state transitions, context sharing, intent routing, and reliable handoffs across vendor and internal LLM frameworks in a distributed conversational environment.
Develop production-grade Python services that bring AI and ML capabilities into customer-facing products. Own well-scoped ML projects from design through delivery, balancing technical trade-offs.
Collaborate with engineering, Conversation Design, CX, and other stakeholders on secure, compliant, scalable, high-performing experiences. Contribute to system design, coding standards, development practices, and design reviews.

Required qualifications

2+ years of professional experience in machine learning and software engineering, including shipping production-grade ML services.
Hands-on experience with modern AI architectures such as LLMs and deep learning; familiarity with tools such as LangGraph, LangSmith, Google ADK, Vertex AI, or AWS Bedrock.
Strong Python proficiency and ability to write clean, maintainable, well-tested production code. Working knowledge of at least one of NLP, information retrieval, computer vision, or statistical modeling.
Ability to document and communicate ML system designs across teams. Responsible use of generative AI with human oversight to achieve measurable improvements in efficiency, cost, and quality.

Location and compensation: Remote in India, with quarterly in-person working sessions. The stated annual base salary is ₹4,408,400—₹4,408,400 INR, excluding potential equity and bonus. India employee benefits include insurance, an employee stock purchase plan, wellness and learning allowances, and leave. Candidates may submit no more than three applications within six months.

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

Skills for this role

Machine LearningSoftware EngineeringPythonLarge Language ModelsDeep LearningLangGraphLangSmithGoogle ADKVertex AIAWS BedrockNatural Language ProcessingInformation RetrievalComputer VisionStatistical ModelingMulti-agent SystemsAI OrchestrationTechnical Design DocumentationSystem DesignGenerative AI

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

Hybrid conversation orchestration
The role centers on a unified layer for state transitions, context sharing, intent routing, and handoffs among vendor AI, internal agents, and people. Prepare to explain how you would design and test those boundaries.
Production-grade ML services
The posting calls for reliable, customer-facing Python services and clean, maintainable, well-tested code. Prepare a design-to-delivery example that makes reliability and testing decisions concrete.
Scoped project ownership and technical judgment
This IC4 role involves driving well-scoped ML projects from design through delivery and balancing technical trade-offs. Prepare to explain how you would define scope, document a design, and justify choices in review.
Secure, measurable support experiences
The team partners with engineering, Conversation Design, and CX on secure, compliant, scalable support; the requirements also emphasize human oversight and measurable efficiency, cost, and quality. Prepare to discuss how those considerations shape a proposed feature and its evaluation.

Likely challenges to think through

Preserving context across handoffs
Likely preparation challenge, not a reported problem: routing a conversation between vendor AI, internal agents, and humans could lose or misinterpret state. Prepare a handoff design that specifies what context carries forward, how transitions are validated, and what happens when routing fails.
Balancing answer quality with operating constraints
Likely preparation challenge: improving support answers may compete with latency, cost, and reliability in a customer-facing service. Prepare to compare orchestration options using explicit quality, efficiency, and cost measures, with human oversight where appropriate.
Reconciling cross-team design constraints
Likely preparation challenge: Conversation Design, CX, and engineering may prioritize different aspects of an experience, while features must meet security, compliance, and performance standards. Prepare to show how a technical design would make trade-offs visible and incorporate review feedback before delivery.

Practice your answers

Technical practice

How would you model state transitions in a conversation spanning vendor AI, internal agents, and a human support agent?

What to coverThe role calls for an orchestration layer that manages state transitions across these participants. Suggested preparation: explain how you would define valid transitions, preserve conversational state, and handle an interrupted or repeated transition.

What context should pass between the Help Center, chatbot, and agent workflows?

What to coverThe posting connects these experiences and explicitly calls for context sharing. Suggested preparation: distinguish information needed to continue a case from information that should be minimized or withheld, and explain how you would keep shared context current.

How would you route a customer request to vendor AI, an internal agent, or a human?

What to coverIntent routing and coordination among those participants are stated duties. Suggested preparation: discuss routing signals, uncertain intent, and the trade-off between faster automated answers and timely human involvement.

What makes a handoff between LLM frameworks reliable?

What to coverThe role specifically emphasizes reliable handoff logic across vendor and internal frameworks. Suggested preparation: describe a handoff contract for state and context, then explain how you would detect and recover from an incomplete handoff.

How would you troubleshoot inconsistent conversation state in a distributed orchestration service?

What to coverThe posting describes a distributed conversational environment and reliable customer-facing services. Suggested preparation: walk through how you would trace state changes across participants, isolate the failure, and prevent a customer or agent from receiving conflicting context.

How would you structure and test a production Python service that exposes ML capabilities to customer support?

What to coverProduction-grade Python services and clean, maintainable, well-tested code are explicit requirements. Suggested preparation: cover service boundaries, tests for routing and failure paths, and how you would balance code clarity with performance.

How would you evaluate an LLM orchestration framework for this role?

What to coverThe posting names LangGraph, LangSmith, Google ADK, Vertex AI, and AWS Bedrock as examples of tools to know. Suggested preparation: compare relevant capabilities such as state handling, observability, integration effort, and portability without assuming the team uses any particular tool.

How could NLP or information retrieval help a support chatbot use Help Center content?

What to coverThe team connects the Help Center with chat experiences, and the qualifications include working knowledge of at least one domain such as NLP or information retrieval. Suggested preparation: explain a plausible approach to finding relevant content and checking whether an answer is supported, while treating that approach as an inference rather than a stated system design.

How would you measure whether a hybrid chatbot improves support outcomes?

What to coverThe posting seeks measurable, scalable products and improvements in efficiency, cost, and quality. Suggested preparation: propose measures for answer usefulness, handoff effectiveness, latency, and cost, and explain how you would spot trade-offs between them.

What would you examine in a security, compliance, and performance review of an AI support feature?

What to coverThe role includes design reviews against security, compliance, and performance standards, as well as responsible generative AI use with human oversight. Suggested preparation: discuss data exposure across handoffs, conditions for human review, and performance testing without assuming specific internal policies.

Behavioural practice

Tell me about a well-scoped ML project you took from design through delivery.

What to coverThe role calls for ownership of well-scoped ML projects and explicit technical trade-offs. Suggested preparation: use a real situation, your design and delivery actions, the trade-off you made, and the observed result.

Describe a time you worked with non-ML stakeholders to improve a user-facing experience.

What to coverThe posting names collaboration with Conversation Design, CX, and engineering teams. Suggested preparation: give a real situation, how you aligned differing needs or explained a technical constraint, and the result.

Give an example of a design review that changed your approach.

What to coverThis role participates in design reviews and contributes to system design and development practices. Suggested preparation: describe the original proposal, feedback you acted on, and the resulting change or outcome.

Tell me about a time you improved the reliability of a production service.

What to coverThe job emphasizes production-grade ML services and reliable customer-facing experiences. Suggested preparation: explain the situation, how you diagnosed and addressed the issue, and the measured or observed result without claiming a particular incident type.

Describe a time you introduced human oversight into an AI-enabled workflow.

What to coverThe posting explicitly calls for responsible generative AI use with human oversight and measurable improvements. Suggested preparation: use a real example to explain when people intervened, what you changed, and what happened to quality, efficiency, or cost.

Sharpen your knowledge

Topics to prepare

Distributed conversation orchestration
The role explicitly covers state transitions, context sharing, intent routing, and handoffs among vendor AI, internal agents, and humans. Revision exercise: draw a conversation state diagram and explain what happens when intent changes or a handoff fails.
LLM architectures and framework boundaries
The posting requires hands-on experience with LLMs and deep learning and familiarity with tools such as LangGraph, LangSmith, Google ADK, Vertex AI, or AWS Bedrock. Revision exercise: compare two approaches to coordinating vendor and internal LLM components, including their context and failure-handling trade-offs; no single listed tool is mandated.
Production ML system trade-offs
The role involves shipping customer-facing ML services and owning well-scoped projects from design through delivery. Revision exercise: review a hypothetical chatbot change and articulate trade-offs among reliability, performance, scalability, and delivery scope.
Responsible AI and measurable outcomes
The posting calls for human oversight, secure and compliant experiences, and measurable improvements in efficiency, cost, and quality. Revision exercise: define when a support interaction should go to a human and propose one measure for each of those three outcomes.
One applicable ML domain
Working knowledge of at least one of NLP, information retrieval, computer vision, or statistical modeling is required. As a preparation suggestion, choose a domain you know and explain one technique and its limitations using a support-chat example; the posting does not require all four.

Skills to revise

Implement a tested Python service
Strong Python and clean, maintainable, well-tested production code are explicit requirements. Revision exercise: write a small service function that accepts conversation context and returns a routing decision, with tests for missing context and invalid inputs.
Exercise handoff logic
Reliable handoffs across vendor and internal LLM frameworks are central to the advertised work. Revision exercise: build a small simulated flow that routes an intent between vendor AI, an internal agent, and a human; test context preservation and failed handoffs. The simulation is a preparation suggestion, not a specified project.
Write and present a system design
The posting requires technical design documents and clear communication across teams, including Conversation Design, CX, and engineering. Revision exercise: draft a one-page orchestration design covering interfaces, state, risks, and trade-offs, then practice explaining it to a non-ML stakeholder.
Evaluate an AI-enabled support change
The role calls for responsible generative AI use and measurable efficiency, cost, and quality improvements. Revision exercise: create a small set of example support conversations, specify human-review points, and describe how you would compare outcomes before and after a routing change.

Your preparation checklist

Prepare a production ML project walkthrough
If you have a relevant shipped service, outline its design-to-delivery timeline, your contributions, Python service architecture, trade-offs, tests, and measured results. The posting asks for production-grade ML services and ownership of well-scoped projects.
Draft a hypothetical orchestration design
Create a one-page design for a support conversation moving among vendor AI, internal agents, and a human agent. Show state transitions, context sharing, intent routing, handoff failures, and recovery; this is practice based on the role, not a stated interview exercise.
Rehearse a Python service review
Prepare a small code example or design sketch with clear interfaces, tests, and failure handling. Be ready to explain how you would maintain reliability in a customer-facing service.
Select focused AI and domain examples
Prepare examples you can accurately discuss involving LLMs or deep learning and at least one listed domain, such as NLP or information retrieval. Explain your actual familiarity with any relevant tools named in the posting rather than claiming experience with all of them.
Define an evaluation and oversight plan
For the hypothetical support design, propose how to check answer quality, handoff reliability, efficiency, and cost, including when a human should review or take over. These are preparation suggestions grounded in the posting’s emphasis on measurable, responsible AI.
Practice communicating a design decision
Prepare a short explanation of one ML systems trade-off for engineers, Conversation Design, and CX stakeholders, including how you would address security, compliance, scalability, and performance in a design review.
Write targeted employer questions
Bring questions about how the team divides responsibilities between vendor AI and its internal multi-agent platform, what context must survive handoffs, and how it measures customer and agent outcomes. The posting describes these goals but does not specify their current implementation.

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…