AI Systems Engineer
Build AI systems that actually run businesses.
What you'd build
Onera builds the engineering layer for companies becoming AI-native. We connect the systems a business already uses, CRM, email, meetings, documents, databases, and turn them into AI-powered workflows that take real actions. Not chatbots. Not demos. Not "AI features" nobody uses.
You'll have real ownership from day one: architecture, engineering standards, AI tooling and evaluation, database design, deployment patterns, observability, testing strategy, internal libraries, and how we turn one-off client work into a platform.
The metric we care about:
Is the next system cheaper, faster, and better to build because we built the previous one?
Every engagement starts with a business problem and a measurable target. You'll get a written spec, the business problem, the workflow being changed, the current baseline, the KPI, the 60-day outcome, and the technical requirements, and turn it into a production system. If the spec or the architecture is wrong, push back. If there's a simpler solution, propose it.
The work:
- Designing multi-step AI workflows that execute reliably in production
- Building agents that interact with real business systems
- Connecting APIs, databases, CRMs, email, calendars, documents, and systems that weren't designed to work together
- Designing Postgres schemas and RLS policies for multi-tenant applications
- Building evaluation systems for LLM-powered workflows
- Creating internal interfaces where humans review, approve, or override AI decisions
- Building ingestion and knowledge systems that make a company's information queryable
- Monitoring production systems and improving their reliability and cost
- Turning successful implementations into reusable infrastructure for the next customer
The stack
Every week: TypeScript, Next.js, Postgres / Supabase, Vercel, Claude Agent SDK, GitHub Actions
Regularly: Redis, Railway, Sentry, Vercel Workflows & Queues, LLM evaluation frameworks
Occasionally: Python, headless browsers, third-party APIs, whatever system a customer happens to run
Postgres and TypeScript are the only hard requirements. We care more about whether you can learn a new system quickly than whether you've memorized a particular framework.
Who fits
- 2-4+ years building software people actually rely on
- Experience running a Next.js or similar application in production
- Strong Postgres fundamentals: queries, indexes, transactions, schema design, ideally RLS
- Experience building something real with an LLM: an agent, pipeline, classifier, extraction system, RAG system, or similar
- A strong instinct for reliability and edge cases
- Works independently without needing a ticket for every decision
- Good written communication and a habit of documenting decisions and assumptions
- Would rather build a system than maintain a ticket queue, solve an ugly integration problem than implement another CRUD screen, design the architecture than wait for someone else to hand it over
- Comfortable being engineer #1: no groomed backlog, no narrow lane, messy existing systems, work reviewed directly by the founders
Bonus points: built LLM evaluations, worked with agent frameworks, built production background jobs or workflow systems, worked with queues and event-driven architectures, integrated messy third-party APIs, worked on multi-tenant SaaS, used Claude Code extensively, contributed to open source, built something of your own.
Who doesn't
- Needs a fully groomed backlog before starting
- Prefers highly specialized roles with narrow ownership
- Only enjoys greenfield development
- Doesn't want to work with messy existing systems
- Dislikes writing things down
- Treats testing and observability as someone else's responsibility
- Wants to experiment with AI but isn't interested in making it reliable enough for production
- Isn't comfortable having their work reviewed by the founders
How we work
Async by default. Onera is based in Tirana and US and works primarily across European hours. A few hours of overlap matter; the rest of the day is yours. We care about output, not online presence.
Autonomy comes with accountability. Anything that reaches production needs CI, smoke tests, monitoring, a runbook, sensible API spend limits, and a clear rollback or failure strategy. "Works locally," "code complete," and "verified in production" are three different statements.
Compensation
Competitive, based on experience and engagement type. Open to both full-time and contract arrangements.
[Apply]
Apply for this role
Send your CV, a link to your GitHub, LinkedIn, or portfolio, and a link to something you built that other people actually relied on. If you've built an AI system, say what it did and how you evaluated whether it worked. One engineering decision you're proud of, explained clearly, beats a paragraph about being passionate about technology. Everything you send is read by the founders.