AI Engineer

Nortiksoftware; United States; Remote; USD 150,000 - 210,000; Full time

Benefits offered by this job

Fully remote across Europe

22 paid vacation days plus tenure days

Private health insurance

Learning budget and conference support

Research time for exploring new models

Top-tier hardware and AI tools

Job summary

Nortiksoftware is seeking an experienced engineer to build AI systems that scale to real production use. You will own end-to-end AI features, from discovery calls to production, designing retrieval strategies, and shipping robust, observable systems with cost controls.

Strong Python and TypeScript skills, plus experience with major model APIs, are essential. You will work on data flow, latency, and failure modes, ensuring guidance for clients and integrating AI into production-grade workflows

Qualifications

• Strong software engineer with a focus on latency, cost, and data flow.
• Experience shipping AI-backed systems with end-to-end ownership.

Responsibilities

• Own AI features end to end from discovery to production.
• Design retrieval strategies and implement agents and LLM-backed workflows.
• Evaluate model performance and communicate results to clients.

Skills

Python

TypeScript

LLM

Retrieval

Observability

Tools

FastAPI

AWS

GCP

Vercel

Postgres

pgvector

Job description

We're the AI engineering partner enterprises bring in when the work has to be right the first time: regulated data, production load, and systems that have been running for a decade before we arrive. You'll ship to production in weeks rather than quarters, own what you build end to end, and work beside a senior team that has already done it at that scale.

Serbia Full-time Mid to Senior Engineering

We're looking for an engineer who builds AI systems rather than demos of them. The distance between a notebook that produces an impressive answer and a system that produces a correct answer ten thousand times a day is most of the job, and it's the part we get hired for. If you find that gap interesting instead of tedious, this is your role.

You don't need a research background or a published paper. You need to be a strong software engineer first, someone who can reason about latency, cost, failure modes, and data flow, with genuine depth in how modern models behave and where they break. We care far more about what you've shipped and what you learned when it went wrong than about which papers you've read.

What will be your responsibilities?

You'll own AI features end to end: sitting in the discovery call where a client describes a business problem, deciding whether a model is even the right answer, then designing, building, and shipping the system that solves it. Our teams are small, so the person who designs the retrieval strategy is the person who implements it and the person who explains to the client why the numbers moved.

Concretely, that means building agents and LLM-backed workflows that behave predictably, designing retrieval over messy client data, writing the evaluations that tell you whether a change actually helped, and taking all of it to production behind proper observability, cost controls, and guardrails. When a system misbehaves in front of real users, you're the one who reads the traces and figures out why.

You'll also do the honest part of AI work: telling a client when a simpler approach beats a model, when their data isn't ready, or when the accuracy they want isn't reachable yet. As you grow into the role you'll set the patterns other engineers build on and help scope new engagements.

This position is made for you, if you:

• Have shipped an LLM-backed system to real users and watched it meet the real world, not just prototyped one.
• Are a strong software engineer before you're an AI engineer: you think about latency, cost, failure modes, and data flow by reflex.
• Measure instead of guessing: you build the eval before you trust the improvement.
• Are comfortable with non-determinism, and can design systems that stay reliable on top of components that aren't.
• Debug from first principles, including through a trace of a model doing something you didn't expect.
• Can say "a model is the wrong tool here" to a client who came asking for AI, and explain why.
• Handle ambiguity well: most engagements start from a business problem, not a spec.
• Communicate clearly in English, in writing especially. We're remote, so decisions live in documents and threads.
• Stay current without chasing every release; you can tell a genuine capability shift from a launch post.
• Give and take direct feedback without it becoming personal.
• Strong Python, and comfort reading and writing TypeScript when the system needs it.
• Hands-on experience with the major model APIs (Anthropic, OpenAI, and the open-weight ecosystem) and their tradeoffs.
• Real depth in one or more of: agentic systems and tool use, RAG and retrieval quality, structured extraction, or fine-tuning.
• Practical grasp of embeddings, chunking, reranking, and vector stores, including when they're the wrong answer.
• Evaluation you actually trust: golden datasets, LLM-as-judge where it fits, regression suites that catch drift.
• Prompt and context engineering as an engineering discipline, versioned and tested like code.
• API and backend fundamentals: FastAPI or similar, queues, streaming, background jobs, sensible data modelling.
• Cloud and deployment: containers, CI/CD, and AWS, GCP, or Vercel.
• Observability for AI systems: tracing, token and cost accounting, and quality monitoring in production.
• Bonus: MCP, model serving and inference optimisation, or classical ML and data engineering experience.

Our projects and stack

Our work spans AI products, fintech, healthcare, logistics, and B2B SaaS, usually as the engineering team a company brings in when AI has to stop being an experiment and start carrying load. Some engagements are a greenfield agent or copilot; others are AI capability threaded into a system that's been running for a decade, where the hard part is the integration and the data, not the model.

Python is the default for AI and data work, with TypeScript, Next.js, and Node for the product surfaces around it. We run the major model providers behind an abstraction we control so a model swap is a config change, Postgres and pgvector or a dedicated vector store for retrieval, and proper tracing and evaluation on everything that reaches production. Infrastructure runs on AWS, GCP, and Vercel with CI/CD from day one. We're deliberate about what we adopt: new tooling has to earn its place in a system somebody will maintain for years.

Benefits that we offer

• Fully remote across Europe, with flexible hours built around a few shared core hours.
• 22 paid vacation days, plus an extra day for every year you spend with us.
• Private health insurance.
• A yearly learning budget for courses, books, and certifications, plus a conference of your choice, on us.
• Paid research time: exploring what the new models can do is work, not overtime.
• Top-tier hardware and a paid subscription to every AI tool worth having.
• Direct mentorship from senior engineers and founders who still write code.
• A referral bonus when you bring us someone great.

Our hiring process

Four steps, two to three weeks end to end, and a human answer at every one of them. You’ll know where you stand the whole way through.

01 Intro call

Thirty minutes with an engineer, not a recruiter. We walk you through what we actually build and who for, you tell us what you want your next two years to look like, and we both decide whether it's worth going further.

02 Technical deep-dive

Ninety minutes on real work. We pull apart a system you've shipped, then work through a problem shaped like the ones we take on. No whiteboard trivia, no algorithm quiz you'd never use again.

03 Meet the team

A conversation with the people you'd sit beside every day, plus the founders. You ask the uncomfortable questions about deadlines, on-call, and disagreement, and get honest answers before you commit to anything.

04 Offer

We move fast here: a decision within a few days of the last call, with the offer, the level, and the reasoning behind both laid out in writing. If it's a no, you get the actual reason.