Architect the orchestration layer that fuses vendor AI and internal multi-agent chat at Coinbase

This is an IC5 ML engineering interview for Coinbase's CX Intelligence team, centered on designing the unified orchestration layer that routes state, context, and intent across vendor chatbots, internal multi-agent systems, and human agents. Expect deep technical design questions on LLM orchestration architecture, production Python services, and cross-team technical leadership rather than pure ML modeling trivia.

Step into this interview

Free · a live voice mock calibrated to this exact role

Practice this interview

What this interview tests

Multi-agent/LLM orchestration architecture (state, context, intent routing)
Hybrid vendor + internal system integration and hand-off design
Production-grade Python service design and testing discipline
Generative AI frameworks (LangGraph, LangSmith, Vertex AI, AWS Bedrock, Google ADK)
Technical leadership: design docs, trade-offs, mentoring, design reviews
Responsible/human-in-the-loop use of generative AI tools

Common question themes

Design an orchestration layer that hands off a conversation between a vendor chatbot and an internal agent mid-session — how do you preserve context and detect failure?

Walk through a production ML/AI service you shipped end-to-end — what broke in production and how did you catch it?

How would you structure intent routing across multiple LLM frameworks with different latency/reliability profiles?

Describe a time you had to make a technical trade-off under cross-functional pressure and how you communicated it to non-technical stakeholders

How do you apply human-in-the-loop practices when using generative AI copilots in your daily workflow?

What does your specialized depth (NLP / IR / CV / stats) bring to a conversational AI system like this one?

View the original posting