Senior AWS Serverless Engineer — Bedrock AgentCore Support Agent Build

What you'll build • A new Bedrock AgentCore agent runtime (the Support Intake Agent) with its own system prompt, memory configuration, and tools. Its job is to gather a fixed set of required fields (issue description, timestamp, reproduction context, frequency) through natural conversation, not a form-like sequence of questions, and to recognize when it already has enough context to skip questions entirely. • Modifications to the existing orchestration Lambda so it reads which agent is active from the session table, routes each incoming message accordingly, and handles the handoff signals both agents emit in their response metadata (handing off to support, handing back to the patient-facing agent, or bypassing to a clinical path if the conversation shifts to something clinical mid-support). • A new DynamoDB table that tracks support session state independently from the main patient session table, including which required fields have been gathered so a session can resume correctly if the patient closes the app mid-conversation and comes back days later. • A new Lambda tool, callable by the Support Intake Agent, that validates the required fields are present, writes the ticket to Aurora PostgreSQL, creates a matching conversation in Intercom through its API, and emits an EventBridge event announcing the ticket was created. This event is a hard boundary: everything downstream of it (ticket classification, routing to a human, resolution) is out of scope for this contract. • Handling for the edge cases that come with a conversational intake flow: a patient abandoning the flow mid-gather, a patient returning to an abandoned ticket later, duplicate-ticket prevention when the same issue is reported twice within 24 hours, and the intake agent resolving something simple (like an order-status lookup) without creating a ticket at all. • Automated tests covering a defined set of routing and handoff scenarios, so the routing logic can be verified before every deploy rather than only in manual testing. • Code and infrastructure that leave clean extension points for a future feedback pipeline that will eventually take resolved tickets and feed them back into agent evaluation. You will not build that pipeline. You will be given the event contract to design around, so it can be added later without touching the code you write now. What a strong submission demonstrates • You can read an existing AWS serverless codebase, understand its conventions, and extend it in that same style rather than introducing a parallel pattern. • You understand the difference between designing a system and following a spec, and you ask sharp clarifying questions about the parts of the spec that are ambiguous rather than guessing. • You've built or maintained something with an LLM agent making autonomous tool calls, and you understand the failure modes that come with that: a model deciding it has enough information when it doesn't, a tool call failing partway through a multi-step action, a model looping instead of terminating. • You write tests that would actually catch a regression in agent behavior, not just tests that exercise the Lambda code paths. Required skills • Production experience with AWS Lambda, API Gateway (REST APIs with streaming Lambda responses), DynamoDB, EventBridge, and Aurora PostgreSQL. • Strong Node.js/TypeScript backend development. • Hands-on experience with AWS Bedrock, Bedrock AgentCore, or a comparable LLM agent orchestration framework (Amazon Bedrock Agents, LangGraph, or similar). • Experience designing event-driven, asynchronous, decoupled systems. • Ability to read and extend an existing codebase quickly and follow established patterns, rather than starting fresh. Preferred • Experience integrating with the Intercom API or a comparable support ticketing platform. • Experience with healthcare or other compliance-sensitive systems, including HIPAA-aware development practices. • Familiarity with AWS CDK for infrastructure as code. Engagement details • Duration: 4 weeks, full-time or near full-time hours. • Start date: immediate. • Working style: async-friendly with structured checkpoints against our internal engineering team. We will provide the architecture specification, an existing reference implementation to pattern-match against, and code review throughout. You will not be building this in isolation. • Definition of done: the Support Intake Agent is deployed to our development environment, the orchestration Lambda correctly routes and hands off between agents, tickets created through the flow are visible in both Aurora and Intercom, the EventBridge event fires correctly, and the automated test suite covering our routing and handoff scenarios passes. • Deliverable: production-ready code, deployed and tested in a development environment, with automated test coverage against our defined routing scenarios.