AI/ML Engineer

About YLAB

We build the AI we would give to our own children.

Children are growing up with AI today. We make sure this AI sparks curiosity instead of addiction, is safe instead of data-hungry, and is built for children instead of adults.

We are YLAB, a small AI lab from Köln-Ehrenfeld. We develop software and hardware in-house, from the language model to the housing. Our first product is Onoo, a screen-free learning companion for children from age four. Onoo asks questions back, lets children guess, and thinks along. It has no camera and no display, shows no advertising, and sells no data.

Why YLAB

Your work ends up in the child's room, not in a dashboard.

Software & Hardware: AI, app, firmware and device come together with us. You see how code becomes something a child picks up and holds.

You work on real open questions, for example how AI understands children's voices and explains the world to five-year-olds.

Tech, pedagogy and research sit at one table with us.

Honesty over glossy finish: we are building much of this for the first time. If that excites you, we should talk.

What we offer

• Responsibility from day one: You get real topics and problems instead of tickets, and your decisions carry weight.

• Top work equipment: You work with the setup and tools you need.

• Flexible further education: Conference, course or workshop – you tell us what moves you forward.

• Hybrid with heart: Remote when you need quiet, and in the office when we tinker together.

About the role

Onoo listens, thinks along and responds in a child-appropriate way. Behind it are speech recognition, a curated knowledge space and a multi-stage safety pipeline. The prototype works, the launch is in 2027. Now we are building the pipeline that can withstand millions of children's questions.

You are our first dedicated AI/ML position, with a seat at the founders' table. You work directly with Lennard (CEO, Product & AI) and help decide what our stack looks like. From model routing to the latency budget on the device.

What to expect: real edge cases such as children's voices, dialects and nonsense questions, along with hard limits on latency, costs and data protection. And the responsibility that every answer is safe before it reaches a child.

How you can help us

• You build and operate our speech pipeline: speech recognition for children's voices, dialogue management, speech output. Fast enough for the living room.

• You orchestrate language models in a provider-agnostic way: routing, prompting, evals, cost control.

• You develop our safety layers further: classifiers, guardrails, multi-stage checks before every answer.

• You build evaluation sets from real children's questions and measure what "well explained" means.

• You bring models into production: EU hosting, monitoring, fallbacks, clean degradation.

• You work closely with research and pedagogy so that insights end up in the product.

• You keep our stack simple. A robust path is better than five half-finished ones.

Get in touch if:

• You have brought 3+ years of ML/AI into production, not just into notebooks.

• You know LLM pipelines: prompting, RAG, evals, latency and cost tuning.

• Audio/speech (ASR/TTS) is not new territory for you. Or you are keen to dive into it.

• Python is your home and TypeScript doesn't scare you.

• You prefer making decisions to waiting for finished specs.

• Data protection and safety are part of the design for you, not a checklist.

• You want your work to arrive in the child's room and not in the backlog.

Probably not a fit if: you only want to train models and production bores you. Or if you only get going once requirements, team and processes are finished.

Find Jobs in Germany on Arbeitnow