AI Engineer

About us

We're NoscAi, a Hamburg-based health-tech company building ClinicOS — a cloud practice management system used by private medical practices across Germany. We handle the unglamorous, high-stakes parts of running a clinic: patient records, scheduling, billing, prescriptions, regulated health-network integrations.

We're a team of 25–30 across Hamburg and Indonesia, growing fast. Our engineering squads are remote-first, English-speaking, and shipping to production every week.

The role

Doctors spend a large share of their day writing. Discharge letters, referral letters, consultation notes — dictated, retyped, reformatted, sent. We're building the AI layer of ClinicOS to take that work off their hands.

But this role is broader than that. Alongside the LLM-powered features, we run our own in-house machine learning system, and you'll be extending it. That means real ML work — data pipelines, training, evaluation, deployment — not just orchestration around a vendor API. The variety here is the point: one week you're tuning a retrieval pipeline for letter generation, the next you're rebuilding how a model gets trained and served inside our own stack.

This is applied work with a hard quality bar. When a model gets a medication name wrong, that's not a bad demo — it's a patient safety issue. We're looking for someone who takes that seriously and builds accordingly.

What you'll do

• Build and improve LLM-powered features for clinical documentation: letter generation, structured extraction from consultations, speech-to-text pipelines
• Extend and maintain our proprietary ML system — data pipelines, model training, deployment, monitoring
• Design agent workflows that call internal tools and APIs reliably, and fail safely when they don't
• Build the evaluation harness: golden datasets, regression suites, and metrics that tell us whether a change actually made things better
• Tune for the real constraints — latency doctors will tolerate, cost per document, and strict data protection requirements
• Work directly with our clinical and product people to turn messy real-world workflows into something a model can handle
• Ship to a production Kubernetes environment alongside our platform engineers

What we're looking for

Must have

• 3+ years of professional engineering experience, with a strong machine learning foundation — you understand what's happening under the hood, not just which API to call
• You have shipped at least one LLM-powered application to production and can talk in detail about what broke, what you measured, and what you'd do differently
• Strong Python; comfortable in a TypeScript/Node codebase (or willing to get there quickly)
• Practical experience with retrieval, prompt engineering, structured output, and model evaluation
• Solid engineering fundamentals: version control, testing, code review, CI/CD
• Professional working English, written and spoken

Nice to have

• Speech/ASR experience (Whisper, diarization, domain adaptation)
• Experience with healthcare, legal, or another domain where accuracy is non-negotiable
• Vertex AI, AWS Bedrock, or comparable managed model platforms
• Fine-tuning, distillation, or self-hosted inference

You do not need German. Our engineering team works in English, and you'll have German-speaking clinical colleagues to work through domain questions with you. Curiosity about the medical domain matters far more than the language.

Read this part before you apply

We're scaling fast, and we're not going to dress it up.

This is a startup in the way people picture startups. Deadlines that don't move. Long hours when something is on fire — and in healthcare software, things catch fire at inconvenient times. Decisions made on incomplete information, because waiting costs more than being wrong. Real stakes: our software sits inside working clinics, and when it breaks, doctors can't treat patients.

If you want a predictable schedule, a narrow scope, and someone else handing you tickets, you will be unhappy here within a month. Please apply somewhere else — we'd genuinely rather you self-select out now than three months in.

If you want the opposite, this is one of the few places where a single engineer still owns an entire domain end to end. You'll join two AI engineers already working on this — a team small enough that you own real surface area from week one, and senior enough that you're not figuring it out alone. What gets built, how it's evaluated, and where it goes next is a conversation you'll be part of, not a roadmap handed to you. That is a hard job and a rare one.

And we want you to own a piece of the outcome, not just the workload. This role is eligible for our VSOP — a virtual stock option plan that gives you a contractual right to a cash payout if NoscAi is acquired or exits, vesting over time. It's not shares and it's not a guarantee; it's a real stake in whether this works. We'll explain exactly how it's structured before you sign anything.

How we work

• Remote-first with real ownership — you're responsible for these features, not executing someone else's tickets
• AI-augmented development is the norm here; Claude Code is our team's primary development tool
• Daily overlap with our Hamburg team (CET) for standups and pairing
• Direct access to the CTO and the product team; short decision paths, no layers to route through
• Long-term collaboration — we're building a permanent Indonesian engineering presence, not filling a gap

How to apply

Send us your CV and a short note on one LLM or ML system you've put in front of real users — what it did, how you knew it was working, and what you'd rebuild today.

Links to code, papers, or side projects are welcome. A cover letter written by an LLM is not.