AI Automation Engineer — D2C Telehealth Platform
Location: Remote (US hours) · Type: Full-time · Team: Engineering
About Us
We're a direct-to-consumer telehealth company operating a high-volume, ad-driven patientacquisition funnel. We're building an agent-first operations layer on top of our platform —retention, support, disputes, lifecycle communications, and monitoring are all beingdesigned to run as orchestrated AI agents with humans in the loop. You'll help take thatarchitecture from design docs to production.
The Role
You'll build, deploy, and operate AI agents and automation across the patient lifecycle. Wehave a formal multi-agent architecture already scoped — spanning retention, win-back,support triage, dispute handling, and lifecycle journeys — and the foundation work(queues, schemas, escalation paths) is being validated by human operators first so agentsinherit a proven structure. Your job is to make those agents real: reliable, observable, safewithin clinical boundaries, and measurably better than the manual baseline.
What You'll Work On
- Multi-agent systems: Implement an orchestrated agent architecture coveringretention (cancellation risk detection, proactive check-ins, win-back), support triage,and dispute handling (triage, evidence compilation, rate monitoring) — each agentwith defined scopes, escalation rules, and human handoff paths.
- Lifecycle automation: Build automated patient journeys on top of our messagingstack (lifecycle/outbound + inbound support platforms) — event-driven flows foronboarding, payment issues, refill reminders, and re-engagement, with agentsprogressively taking over from manual playbooks.
- Agent tooling & integration (MCP): Build the tool layer agents operate through —structured access to orders, subscriptions, payment state, shipping data, andconversation history — using MCP-style tool definitions with strict permissioningaround what agents can read vs. act on.
- Monitoring & anomaly-detection agents: Automated systems that watch our funnelin real time — synthetic checks, client-side degradation signals, and reconciliationagents that verify every submission is processed and reported correctly end to end.
- Observability & evals: Instrument agent behavior (traces, decision logs, escalationrates), define success metrics per agent, and build the eval loops that gate an agent'spromotion from shadow mode to autonomous operation.
- Clinical safety boundaries: Enforce hard lines around what automation may never do— anything clinical routes to named humans, no exceptions — and build theguardrails that make that enforceable in code, not policy docs.
What We're Looking For
- 3+ years of software engineering experience (TypeScript/Node.js and/or Python) withproduction LLM-based systems
- Hands-on experience building agents or LLM workflows — orchestration, tool use /function calling, structured outputs, prompt and context management
- Experience integrating messaging/support platforms (e.g., Customer.io, Intercom,Zendesk or similar) via APIs and webhooks
- Solid SQL/Postgres skills and comfort with event-driven architectures
- A strong instinct for guardrails: permissioning, human-in-the-loop design, andknowing when an agent should escalate rather than act
- Care around handling sensitive data; HIPAA-regulated environment experience is astrong plus
Nice to Have
- Experience with MCP (Model Context Protocol) or building tool layers for agents
- Observability for LLM systems (OpenTelemetry traces, eval frameworks, decisionlogging)
- Experience in telehealth, healthcare, or another regulated D2C vertical
- Full-stack chops (React) for building internal agent dashboards and operator UIs
How We Work
Small team, high ownership. The agent architecture is documented in detail — definedagents, data flows, and safety boundaries — and rollout is deliberately staged: manualoperations first, then agents in shadow mode, then autonomy where the numbers justify it.You'll ship in tight loops with direct access to leadership.
To apply: Send your resume and a short note on an agent, automation, or LLM systemyou've shipped to production — what it did, how you kept it safe, and how you knew itworked.
Work Location: Remote
