Home › Civil Engineer jobs in Denver, CO › A2Z Sync

Senior Engineer, AI/ML

A2Z Sync · Denver, Colorado · Posted November 18, 2026 (1 days ago)

Apply on A2Z Sync’s siteFill the application with AutoFillSee your fit for this role

Job description

Why This Role Exists We operate a multi-tenant automotive SaaS platform serving thousands of dealer groups across the United States, running on a native AWS backend: Lambda, EventBridge, DynamoDB, S3, Step Functions. That platform works. Now we are making it think: production agentic AI systems built code-first and deployed on AWS Bedrock AgentCore, connected to real systems, making decisions and executing workflows against real dealer data and real money. This is a hands-on builder role. Expect the large majority of your time in code. But we are not hiring a pair of hands. Every agent you ship makes decisions that touch dealer revenue, so we need an engineer who understands the business problem before writing the technical solution: someone who asks what a workflow is worth, weighs build cost against dealer impact, and knows when the right answer is a simpler tool, or no agent at all. Scope & Scale 5,000+ destination dealer tenants, each with isolated databases and per-tenant configuration. Agent workflows that touch production transaction data: real dealer money, not a sandbox. Tens of thousands of API requests across REST, SOAP, and event-driven integration surfaces that your agents' tools will call into. A growing portfolio of production agents, each with its own cost, latency, and evaluation budget. What You'll Own Production agents on AgentCore. Agent logic, tool orchestration, and multi-agent workflows built in Strands or LangGraph and deployed on AgentCore Runtime with Memory, Identity, and Gateway. Tool interfaces. Framework-agnostic Python tools, MCP servers, and AgentCore Gateway targets that let agents safely call production APIs. Evaluation harnesses. Offline and online eval pipelines and LLM-as-judge CI/CD gates that catch regressions before dealers do. Guardrails and safe failure. Approval gates and rollback paths that determine whether an agent fails safe or fails loud. Cost and latency management. Per-agent cost tracking and multi-model routing so each workflow earns more than it costs. Production observability. AgentCore session tracing, OpenTelemetry, and CloudWatch dashboards for every agent you ship. You work within the platform's established architectural patterns, and produce deliverables using agentic frameworks, AWS serverless core, MCP and tool design, and observability without someone looking over your shoulder. Prompt engineering and tool-use design are core engineering disciplines here, not an afterthought. First 90 Days / First Year First 90 Days: Ramp on the platform architecture, existing agents, and integration surfaces, and on the dealer workflows behind them, so you know what the agents are for, not just how they run. Ship your first tool integration (MCP server or Gateway target) to production in month one. By month three, own end-to-end agent workflows from design through production on AgentCore: Strands agent logic, tool interfaces, evaluation harness with CI gates, and guardrails. First Year: Running autonomously from month four: shipping new agents and tools with minimal oversight, with observability and cost tracking in place for everything you own. A portfolio of production agents in your name, each with an eval harness, guardrails, and a cost and latency budget it stays inside. At least one case where you argued for the simpler tool, or no agent at all, and were right. Hands-On Expectations The large majority of your time is in code: agents, tools, evals, and the observability around them. The rest is design within the platform's established patterns and working with the Tech Leads whose domains your agents touch. You measure your week in shipped code. Tech & Tools AI and agents: Strands Agents SDK on AWS Bedrock AgentCore (Runtime, Gateway, Memory, Identity, Observability); MCP servers; Bedrock Guardrails; Anthropic models (Haiku, Sonnet, Opus). AWS: Lambda, EventBridge, DynamoDB, S3, Step Functions, ECS Fargate, Aurora, API Gateway, CDK. Languages: Python primary, Java (Spring Boot) secondary. Observability: AgentCore session tracing, OpenTelemetry, CloudWatch. How You'll Use AI This is not a "we are AI-curious" company. We use Claude every day, and we expect you to as well. Concrete examples for this role: Build the production agents themselves: Anthropic models on Bedrock AgentCore, orchestrated in Strands, calling production APIs through MCP servers and Gateway targets. Write eval fixtures and LLM-as-judge CI gates so every agent is tested against what good looks like before a dealer sees it. Use Claude as your pair for agent code, tool schemas, and prompt iteration, with the agentic development harness as your default workflow. Instrument every agent for cost and latency and use multi-model routing so the workflow earns more than it costs. Decide, with evidence, when an agent is the wrong answer and a deterministic tool is the right one. You will not be replaced by AI. You will be the engineer who builds the agents that do real work for real dealers. Role Specifics Department: Engineering Reports To: Head of Data Platform Location: Denver, CO, hybrid (2 days/week in office) or Remote (US, outside the Denver market) Employment Type: Full-Time

Posting q65qDdXbqqC9Di3nBDtyhs on workable. Source: the employer’s listing. Valid through January 2, 2027 unless closed earlier.