Automation Engineer for n8n, GHL, Backend

ScaleAt runs AI-driven growth automation for med spas and aesthetic clinics. We sell a written performance guarantee, which means our backend cannot fail silently — a missed call or a broken retention sequence isn't a bug, it's a refund. We're building the full automation infrastructure (four "pillars": local-SEO/Maps, an AI voice front desk, an AI chat agent, and a patient-retention engine) on a self-hosted n8n + GHL + PostgreSQL stack. We already have a detailed, day-by-day build plan and an AI coding agent producing the work. We need an engineer who can take that plan and make it real, correct, and bulletproof — and keep it running as we add clients. This is not a drag-and-drop no-code role. The hard parts are backend correctness: idempotency, webhook signature verification, concurrency control, retries, and "does this actually still work today" monitoring. What you'll own Build and harden n8n workflows in queue mode (Redis), backed by PostgreSQL — not toy automations, production workflows with retries, backoff, and a central error-handling layer. Wire integrations over REST APIs and webhooks: GoHighLevel (GHL), Retell AI (voice), SearchAtlas, DataForSEO, Anthropic Claude API, Slack/Twilio. Make every handler idempotent and every inbound webhook signature-verified before it acts. Implement concurrency control (e.g. Postgres advisory locks) so simultaneous events can't double-book a calendar. Own the reliability layer: error-trapping, alerting, self-audits, daily/weekly health checks, Postgres backups, git-versioned workflow exports. Work staging-first: test against a staging environment and force failure paths before anything touches a live client. Drive an AI coding agent (Claude Code / Codex) to produce most of the code/config, then review, correct, and integrate it. Comfort directing and sanity-checking AI output is a core part of this job, not a nice-to-have. Must-have 2+ years building real automation/integration workflows (n8n strongly preferred; Make/Zapier + scripting acceptable if the backend depth is there). Solid PostgreSQL: schema design, upserts, indexes, advisory locks, basic query tuning. Genuine understanding of APIs and webhooks: auth types (OAuth2, bearer, basic auth), retries/backoff, rate limits, and verifying webhook signatures (HMAC / Ed25519). A working grasp of idempotency and race conditions — you can explain why "check availability + book" should be two steps, not one. Comfortable on the command line, with git, and with a Linux VPS (Docker a plus). Hands-on with AI coding tools (Claude Code, Codex, Cursor) and able to judge when their output is wrong. Clear written English and the discipline to document what you build for a non-technical founder. Strong plus GoHighLevel (GHL) experience — SaaS mode, snapshots, custom fields, Conversation AI, calendars. Retell AI or other voice-AI / telephony integration experience. Healthcare-adjacent or HIPAA-aware work (we handle patient data; care here matters). Local SEO tooling (SearchAtlas, DataForSEO) and prompt design for content generation. How we work The build follows a written, day-by-day plan with explicit "definition of done" checkpoints per phase. Reliability is a first-class deliverable. "It worked once" is not done; "it survives a forced failure and tells us when it breaks" is done. Nothing goes to production without passing staging tests and (for high-risk workflows) a second-person review. Paid trial task (how we hire) Instead of a long interview, we pay for a short, real task. Build a single n8n workflow that: Receives an inbound webhook and verifies its signature before doing anything. Upserts a row in PostgreSQL idempotently (running it twice with the same payload must not create duplicates). Calls an external HTTP endpoint with 3 retries and exponential backoff. On final failure, routes the error to a central error workflow that logs to Postgres and posts to Slack. Deliver: the workflow JSON export + a 5-minute screen recording walking through how it handles (a) a duplicate webhook and (b) a forced failure. We pay for your time on this regardless of outcome.