Staff Machine Learning Engineer, AI Agent Platform

GEICO•Bethesda, MD

4d•$115,000 - $260,000

About The Position

The GEICO AI Agent Platform team is seeking an exceptional Staff ML Engineer to build the next generation enterprise AI Agent OS and SDKs. You will design, implement, and maintain scalable backend systems that enable business, product, and engineering teams to build, test, and deploy their own AI agents & workflows. In 2026, the agentic AI landscape is maturing rapidly — with standardized protocols (MCP, A2A), AI agent skill ecosystems, harness engineering, context engineering, and governance-first design becoming table stakes. You will help GEICO stay at the forefront. The candidate must have excellent communication skills and a proven track record of delivering business value via technical excellence.

Requirements

Bachelor's in CS, Engineering, or related field; advanced degree highly desirable.
6+ years designing, implementing, and maintaining multi-tenant AI/ML systems in production.
6+ years with cloud platforms (Azure, AWS) and backend systems (Kubernetes, Temporal, OpenSearch, PostgreSQL, Redis, Neo4j).
Deep understanding of Docker, Prometheus, and OpenTelemetry.
Deep proficiency in Python, Java, or Go.
Extra credit for effectively leveraging AI coding tools (Cursor, Claude Code, GitHub Copilot).
Proficiency in AI/ML and agentic frameworks (TensorFlow, PyTorch, LangGraph, CrewAI, AutoGen).
Demonstrated track record mentoring engineers and leading technical initiatives.
Excellent communication across diverse seniority levels and professional backgrounds.

Nice To Haves

Experience with harness engineering concepts and practices such as tool dispatch, error recovery, session state, permissions, sub-agent coordination, planning & reasoning w. feedback loops, etc.
Experience designing AI agent skill systems — reusable capability packages, skill registries/marketplaces with discovery, versioning, security vetting, and governance controls.
Hands-on experience with MCP (server development, registries) and A2A (AI agent card discovery, task delegation).
Experience with LLM observability (LangSmith, Langfuse, Arize Phoenix) and guardrail systems (prompt injection defense, PII scanning, skill-level security auditing).
Experience with multi-agent orchestration, both open-source (Llama, Qwen, Mistral) and proprietary (GPT, Claude) LLMs, and no-code/low-code AI agent development environments.

Responsibilities

Architect scalable multi-tenant backend systems for AI agent workflows — including AI agent configuration, evaluation, synthetic data generation, workflow simulation & evaluation, MCP server registry, A2A communication infrastructure, and guardrail enforcement layers using AKS, FastAPI, etc.
Build an enterprise AI agent skill ecosystem — a platform for authoring, publishing, discovering, versioning, and governing reusable skill packages that encode domain expertise into portable modules.
Implement an internal skill marketplace with search/discovery, quality scoring, security vetting pipelines, approval workflows, and progressive disclosure loading.
Implement production-grade AI agent harnesses — the non-model infrastructure (tool dispatch, context management, error recovery/self-healing, session state, sub-agent coordination) that makes AI agents reliable for long-running tasks.
Design feedforward guides (linters, type checkers, architecture constraints) and feedback sensors (test execution, LLM-as-judge, semantic analysis) mixing computational and inferential controls.
Build and optimize context engineering systems — memory hierarchies (short-term, working, long-term), RAG pipelines, scratchpads, context compaction/summarization, and dynamic skill/tool loading — ensuring AI agents receive the right information at the right time while minimizing token waste.
Develop observability frameworks (OpenTelemetry, distributed tracing) with LLM-specific telemetry: token usage, latency profiling, hallucination detection, AI agent behavior auditing, and skill execution monitoring.
Design layered guardrail architectures (input validation, prompt injection defense, PII detection, output verification) with parallelized enforcement for minimal latency impact.
Implement skill-level governance: security vetting for hidden payloads, credential theft, and data exfiltration risks; authoring standards; conflict resolution; version management; and deprecation workflows.
Act as tech lead for a sub-team, setting direction and ensuring consistency in design principles.
Provide hands-on mentorship during design reviews, code assessments, and performance tuning.
Establish engineering standards for ML infrastructure, harness engineering patterns, skill authoring, and deployment practices.
Create documentation, runbooks, and training on platform capabilities.
Collaborate cross-functionally with data scientists, engineers, and product teams.
Translate complex technical concepts for diverse stakeholders.