Senior Applied AI Enginner, AI Platform

About Eddy at Headway:

Eddy is Headway's in-house AI agent platform and the highest-leverage internal product we've shipped. It gives every Headwegian an agent harness pointed at Headway's own systems: an isolated sandbox per conversation, tool access to the third parties we depend on, and the ability to turn any of it into something durable, like a shareable interactive app, a workflow, a scheduled agent, a Slack teammate, or a webhook-triggered automation. Today 850+ Headwegians use Eddy daily, and 40k+ scheduled agents and workflows run every week with no one in the loop.

We sit in Engineering Foundations and own all of it: the chat product, the agent loop, the sandboxed runtime, the tool and integration layer, the artifact and app platform, the workflow engine, the eval harness, and the safety and access-control model that lets this touch real production and healthcare data without causing incidents. We're a small team with a very large surface area.

About this role:

Self-serve gets a function most of the way. It doesn't get them all the way.

Legal still reads every provider contract and payer amendment by hand, because what would help isn't a chat window. It's a system that knows what a redline means here, what gets escalated, and what our position was the last four times. Credentialing, claims exceptions, and CX triage each have their own version: work that is obviously automatable but full of enough domain-specific detail that nobody outside the function can build it, and nobody inside can build it alone.

That's this role. You'll embed with one function at a time and build them a real product on top of Eddy. Then you'll do the harder half: work out which parts are specific to that function and which are primitives that belong in the platform, and land the primitives in Eddy so the next function gets them for free. Most quarters should end with a function measurably better off and a new capability in the platform.

What you'll do:

- Find the real bottleneck. Sit with a function long enough to see the work, and define success in their terms: turnaround time, escalation volume, error rate, headcount avoided.

- Build the vertical. Skills, MCP integrations against their systems of record, subagents, scheduled and event-driven agents, artifact apps, and whatever product surface the workflow needs. You own it end to end.

- Ship in days, harden over weeks. Get a rough v1 in front of real users fast, then do the unglamorous work of making it dependable.

- Own it in production. Evals, tracing, cost and latency, failure modes. If a function depends on it, you're on the hook for it.

- Extract the primitives. Land the generalizable half in Eddy as a first-class capability, with tests and docs.

- Kill work that shouldn't exist. Some requests are better answered by a workaround, a config change, a process change, or nothing. Finding those is worth as much as building.

- Design for PHI from the start. Eddy operates inside HIPAA.

This is not solutions or sales engineering. There's no deal cycle, and your users are colleagues. It's not a pure platform role either: you'll spend real time inside other teams' problems.

Tools we use:

Languages: Python 3, TypeScript

Frontend: React 19, React Router 7, Tailwind, Radix, TanStack Query

Backend & Storage: FastAPI, SQLAlchemy 2.0 (async), Alembic, Postgres, Redis, DynamoDB, Snowflake

AI: Claude Agent SDK, Anthropic API, Bedrock, LangSmith

Infrastructure: AWS, EKS (a hardened pod per session), EC2 + CodeDeploy blue/green, Temporal, Kafka, Cloudflare

Ops: Datadog, Sentry, PagerDuty, GitHub Actions

AI tools: Cursor, Claude Code, and Eddy itself. We build Eddy with Eddy.

Who you are:

- An end-to-end product builder who moves between a React surface, a FastAPI service, and a container image without a handoff.

- Serious about AI as engineering: tool use, agent loops, evals, context management, and cost and latency at volume.

- Someone who likes users you can see. You've gravitated toward internal platform or developer tooling work, or taught something and enjoyed it.

- Able to hold a scope conversation: push back, renegotiate, say no, and leave people wanting to work with you again.

- A platform thinker who starts specific, and has the judgment to tell which half of a build is which.

- Comfortable owning outcomes with no spec and no backlog.

- Guardrails by instinct, and precise in communication with people who don't write code.

Experience we're seeking:

- 5+ years shipping production software end to end.

- Has built and shipped LLM/agent-powered products to real users.

- Experience as a forward deployed engineer, solutions engineer, internal platform engineer, or founder is a plus, as is work on agent frameworks or developer platforms. Healthcare experience is not required, but you should be able to dive into an exception-heavy domain and build a working mental model fast, several times a year.

You'll love this role if you want to:

- See a real business problem through to a shipped product your colleagues depend on.

- Shape an AI platform from the outside in.

- Have the most cross-functional engineering role at Headway.