Senior Staff Machine Learning Engineer, AI Agent Platform

GEICO GEICO · Insurance · New York, NY +3

Senior Staff ML Engineer to lead the technical vision and architecture for GEICO's AI agent platform. This role involves designing and building multi-tenant services for creating, testing, deploying, and hosting LLM-based AI agents, including orchestration, interoperability, skill ecosystems, harness engineering, context management, and safety guardrails. The role requires extensive experience in AI/ML platforms, LLM systems, agentic AI, and SDLC management.

What you'd actually do

  1. Define the long-term technical strategy for GEICO's AI agent platform — including multi-agent orchestration, AI agent lifecycle management, evaluation frameworks, skill registries and marketplace, and workflow orchestration.
  2. Architect an enterprise skill ecosystem — reusable capability packages that encode domain expertise and workflows into portable, discoverable modules. Build and govern an internal skill marketplace with versioning, security vetting, approval workflows, progressive disclosure loading, and usage analytics.
  3. Lead design of production-grade AI agent harnesses (tool dispatch, context management, error recovery, session state, fine-grained Authn/AuthZ) that makes AI agents reliable for long-running workflows. Apply feedforward guides (linters, architecture constraints, spec-driven validation) and feedback sensors (test execution, LLM-as-judge) mixing computational and inferential controls. Design context engineering systems that treat the LLM context window as a managed resource — memory hierarchies, RAG pipelines, context compaction, scratchpads, and dynamic skill/tool loading.
  4. Own high-performance platform components powering end-to-end agentic workflows: MCP server/registry management, A2A communication infrastructure, prompt management, workflow orchestration, guardrail enforcement, and observability pipelines.
  5. Establish AI agent governance frameworks including bounded autonomy, human-in-the-loop escalation, audit trails, prompt guardrails, and RBAC/ABAC access controls. Extend governance to skill-level security — vetting published skills for hidden payloads, injection vectors, and data exfiltration risks.

Skills

Required

  • 8+ years of professional software development experience with at least two languages (Java, C++, Python, Go, or C#).
  • 6+ years designing and building AI/ML platforms using open-source/cloud-agnostic components (Elasticsearch, Qdrant, Kafka, PostgreSQL, MongoDB, Spark, Ray, Temporal, Redis, Neo4j, etc.).
  • 5+ years managing end-to-end SDLCs (CI/CD, Kubernetes, testing, monitoring, production support).
  • 4+ years building training, fine-tuning, and inferencing systems for LLMs, especially on GPU infrastructure.
  • 3+ years designing and operating multi-agent or agentic AI systems in production.
  • Strong understanding of context engineering — memory architectures, RAG, context compaction, and dynamic information management for LLMs.
  • Demonstrated track record leading technical initiatives, setting architectural direction, and mentoring across teams.
  • Bachelor's degree in CS, Engineering, or related field

Nice to have

  • 6+ years with cloud providers (Azure, AWS), including container orchestration and GPU compute.
  • 3+ years building agentic workflows with open-source and proprietary LLMs (Llama, Qwen, Claude, Gpt, etc.).
  • Hands-on experience with MCP and A2A protocols — MCP server development, AI agent card discovery, task delegation patterns.
  • Experience with harness engineering. (tool dispatch, error recovery, session state, sub-agent coordination, planning & reasoning)
  • Experience designing AI agent skill systems: building and governing reusable skill packages, skill marketplaces with discovery, versioning, security vetting, and progressive disclosure.
  • Experience with context engineering at scale: memory hierarchies, RAG optimization, compaction/summarization, state isolation, etc.
  • Experience with multi-agent orchestration frameworks (LangGraph, AutoGen, CrewAI).
  • Experience with LLM observability & evaluation platforms (LangSmith, Arize Phoenix, Langfuse).
  • Experience building guardrail systems (prompt injection defense, PII detection, skill-level security auditing).
  • Understanding of AI safety, model governance, and regulatory compliance in regulated industries.
  • advanced degree highly desirable

What the JD emphasized

  • multi-agent orchestration
  • AI agent skill ecosystem
  • production-grade AI agent harnesses
  • context engineering systems
  • AI agent governance frameworks
  • multi-agent or agentic AI systems in production
  • context engineering — memory architectures, RAG, context compaction, and dynamic information management for LLMs
  • agentic workflows
  • AI safety, model governance, and regulatory compliance in regulated industries

Other signals

  • AI Agent Platform
  • multi-agent orchestration
  • enterprise scale
  • LLM-based AI agents