The opportunity
Ready to do the most impactful work of your career? At Coinbase , we are uncompromising on our mission to increase economic freedom.
What you'll do
Architect and build production Agentic AI systems for compliance automation,: including LLM-powered agents, orchestration pipelines, and decision support workflows that replace manual compliance processes
Own the ML technical direction for compliance agents, driving model: selection, prompt engineering, grounding strategies, and guardrail design that ensure agent outputs are accurate, auditable, and regulatorily defensible
Design and implement evaluation and measurement frameworks that quantify: agent performance, detect hallucinations and drift, and provide the auditability compliance teams and regulators require
Partner with Compliance, Legal, and Operations stakeholders to translate: complex regulatory workflows into automatable technical specifications, identifying which processes are candidates for full automation vs. human-in-the-loop augmentation
Mentor engineers on ML system design, LLM application patterns, and: responsible AI practices, raising the team's technical bar for production AI systems
+ years in machine learning and software engineering, with demonstrated: experience shipping LLM-based or Agentic AI applications to production
What they're looking for
- Deep expertise in LLM application development, including prompt engineering,: retrieval-augmented generation (RAG), grounding techniques, hallucination reduction, and agent orchestration frameworks
- Proven track record building evaluation and measurement systems for ML models: in production, including monitoring for drift, accuracy degradation, and output quality at scale
- Strong Python proficiency with hands-on experience in GenAI frameworks and ML: infrastructure (e.g., LangChain, vector databases, Airflow, Spark, or equivalent orchestration and data processing tools)
- Experience in compliance, regulatory technology, financial crimes, or: similarly regulated domains where auditability, explainability, and correctness of automated decisions are critical