What you will do

Design and implement automated evaluation frameworks for LangGraph-based agent workflows and orchestration pipelines.

Develop build-time evaluation suites covering agent behavior, tool selection accuracy, reasoning quality, and final output quality.

Create evaluation methodologies combining deterministic grading approaches with LLM-as-judge evaluation techniques.

Build test harnesses for LangGraph graphs, nodes, state transitions, and agent execution paths.

Design multi-turn conversation simulations to validate context retention, memory utilization, and workflow consistency.

Define and maintain reliability metrics, including pass@k and pass^k methodologies, to measure both success rates and behavioral consistency.

Implement CI/CD deployment gates that automatically block releases when evaluation thresholds are not met.

Develop validation processes for staging environments, shadow-mode comparisons, and controlled production rollouts.

Integrate AWS AgentCore Evaluations (on-demand and online evaluation modes) into continuous testing and quality assurance workflows.

Transform production incidents, failures, and unexpected agent behavior into regression test cases.

Establish evaluation metrics, quality benchmarks, and acceptance criteria for agent releases.

Collaborate with AI engineers, platform teams, and product stakeholders to improve agent reliability and performance.

Define observability and feedback mechanisms that connect production behavior with test framework improvements.

Produce documentation for evaluation strategies, scoring methodologies, deployment gates, and quality standards.

What you need for this

Skills:

• Agentic AI system architecture (agent orchestration, multi-agent delegation, tool use, memory/state management)

• Agent orchestration frameworks (LangGraph, Strands, AutoGen, CrewAI, or comparable — production-level depth in at least one)

• Agent-to-tool and agent-to-agent integration patterns (function calling, MCP, A2A, or equivalent protocols)

• LLM application architecture (RAG pipelines, prompt/context engineering, model routing, evaluation of LLM outputs)

• Agent governance and runtime controls (authorization/policy enforcement for autonomous actions, identity for non-human actors)

Experience:

• 6+ years software architecture experience, including 2+ years specifically architecting LLM-powered or agentic AI systems in production

• Hands-on production experience with at least one agent orchestration framework (LangGraph, Strands, AutoGen, CrewAI, or equivalent), owning architecture decisions, not just implementation

• Experience designing agent-to-tool integration at a platform level (a tool/integration layer consumed by multiple agents or teams, not a single agent’s custom tool wiring)

• Experience architecting at least two of the following for agentic systems: runtime/execution environment, tool/gateway integration, identity and policy enforcement, observability and evaluation

• Demonstrated experience presenting architecture for governance or technical review (ARB, design review board, or equivalent)

Nice to have:

• Hands-on experience with AWS Bedrock AgentCore (Runtime, Gateway, Registry, Policy, Evaluation, Observability)

• Experience with policy-as-code frameworks for agent authorization (Cedar, OPA, or similar) • Familiarity with MCP and A2A protocol specifications

• Prior architecture role spanning multiple platform domains in an enterprise, regulated, or highly governed environment

• Experience with agent identity standards (e.g., non-human identity federation, MS Entra Agent ID or equivalent)

• Observability and evaluation architecture specific to agentic systems (tracing multi-step agent reasoning, quality/safety monitoring)

• Cloud architecture supporting managed agent runtimes (AWS preferred — networking, IAM, managed compute)

Desired Candidate Profile