Why we're hiring
AMPD is shipping at speed (~800 commits / month, 13 PRs merged today alone) on a small
team currently drive all engineering decisions, with Claude Code agents drafting and reviewing PRs in worktrees. We need a senior engineer who can plug into this workflow on day one, not someone who needs to learn agentic tooling from scratch.
How we actually work (read this first, it's the most important section)
We are an agent-augmented team. Engineering is a human + Claude Code partnership, not a
human-only job.
Specifically:
- Telegram is the ops surface. The AMPD OPS group chat is where bugs are reported, audits are run, deploys are announced, and decisions get made. – Claude Code reads/writes Telegram via plugin. You'll be expected to participate in this channel, directing agents, reviewing screenshots from CEO, and dispatching fixes.
- Claude Code (or equivalent agentic CLI) is your IDE. Most non-trivial work runs through worktree-isolated background agents. You scope an issue, dispatch an agent in an isolated git worktree, the agent drafts a PR, you (or Claude Code itself, under merge- authority gates) review and ship. We have 31+ active worktree agents in the repo at any time, managing parallelism without stepping on each other is a core skill.
Skills + plugins:
- We use Claude Code skills (e.g. /triage-labels, /grill-with-docs) and MCP plugins (Telegram, GitHub). You should know how to author and consume them.
- GitOps everywhere. Push to main → CI builds → ArgoCD auto-promotes through dev → stage → prod (Hetzner k3s, prod is ampdaxo.com). Manual prod intervention should be rare. No-synthetics rule. AMPD's core promise: real data or empty state, never fabricated. If you write a heuristic fallback, it must be removed before shipping. Surgical-edits rule. No speculative abstractions, no opportunistic refactors. Touch only what the task requires.
The codebase you'll work in:
Backend: Express.js + TypeScript strict, ~140 source files, 65 services, 13 route files, ~1930 Jest tests. Prisma + PostgreSQL (16 models, all CUID, cascade-on-delete). 6 LLM providers (OpenAI, Anthropic, Perplexity, Gemini, Groq, OpenRouter) wired through provider adapters with caching and rate-limiting.
Frontend: Next.js 14 App Router + TypeScript strict, ~313 source files, NextAuth v4 (JWT), Playwright E2E, vitest unit. Tailwind + shadcn/ui.
Infra: k3s on Hetzner, ArgoCD, GitHub Actions (build-only, local pre-push hook runs
lint/build/test), Sentry, Airwallex billing.
Velocity: ~800 commits / 30 days. Small team, high cadence.
Architecture details: docs/ARCHITECTURE.md. Agent workflow: docs/agents/. Domain ADRs:
docs/adr/.
What you'll own:
- GEO/AEO scoring engine. AMPD's three proprietary metrics (CPS, CFI, SAV) plus E-E-A-T, Topic Authority, Semantic Relevance, Content Freshness analysers. You'll improve and extend these.
- AI provider integrations. Adding new platforms, tightening prompts, managing rate limits and Groq credits.
- Audit pipeline. End-to-end from POST /api/scoring through provider calls → metric derivation → AnalysisDerivedMetrics row → dashboard render. Real production data, real money on the line.
- Trust + transparency. When the dashboard shows a number, that number must be defensible. You'll audit your own work the way the CEo audits ours.
What we need from you:
Required:
- 5+ years backend or full-stack TypeScript / Node.js shipping production systems.
- Demonstrated agentic coding experience in Claude Code, Cursor, Aider, or similar. “I've used Copilot” is not the same thing. We want examples of you running multi-agent workflows, dispatching parallel work, managing worktrees / branches, reviewing agent- produced PRs.
- Comfort with prompt engineering as a first-class skill, you should be able to debug an LLM's output the way you debug code.
- Real Postgres / Prisma experience (we have schema migrations going to prod weekly).
- Git fluency including worktrees, rebase, conflict resolution under pressure.
- Excellent written communication, you'll spend significant time directing agents and writing issues. Vague prompts produce vague PRs.
Strongly preferred:
- Experience with Next.js 14 App Router + NextAuth.
- Working knowledge of LLM API mechanics (token economics, streaming, context
windows, retry strategies).
- E2E testing with Playwright.
- GitOps / ArgoCD operations experience.
- Domain interest in GEO/AEO (“how do brands win on ChatGPT/Claude/Perplexity/Gemini?”), this is the entire premise of AMPD.
How you'll be evaluated:
1. Async take-home (paid): dispatch Claude Code (or your tool of choice) to fix a real open issue from the public repo's queue, draft the PR, and walk us through your reasoning +
agent-management decisions. Quality of your prompts and reviews matters as much as
the diff.
2. Pair session over Telegram: CEO and CTO hand you a screenshot of a dashboard anomaly. You scope the issue, dispatch an agent, narrate decisions, review the result.
3. Reference + judgement check. Two prior engineering managers or peers.
Working arrangement: Remote, async-first. Telegram + GitHub for sync; deep work on your own clock.
Compensation: competitive senior engineer salary or day rate, equity discussion for permanent. Send your range.
Location: Anywhere in the world but need to be available for UK core work day hours, strongly preferred for overlap with CEO and CTO.
How to apply: Email hello@ampdaxo.com with:
1. A short note on your most interesting agent-driven engineering project.
2. A link to a PR you shipped end-to-end with significant agent involvement, with
commentary.
3. Your range and earliest start date.
We read everything. Bots / form letters skip to the bin.
4. Your CV.
…