AI Researcher

Reson8; Amsterdam; Hybrid; EUR 90,000 - 150,000; Full time

Benefits offered by this job

Salary competitive with the top of the

Equity as an early employee

25 days vacation

Job summary

Reson8 is building advanced speech-to-text systems for European languages from Amsterdam. You will train and fine-tune models, focusing on domain-specific vocabulary, and own the full experimentation loop from research to deployment.

You’ll work closely with founders with engineering backgrounds to push model accuracy, maintain infrastructure, and stay at the cutting edge of NeMo and TensorRT with multi-GPU training on our NL cluster.

Qualifications

• Hands-on experience training and evaluating deep learning models (speech/NLP/seq-to-seq).
• Proficiency in Python and PyTorch; practical model training beyond notebooks.

Responsibilities

• Develop and improve Reson8's speech-to-text models across 24+ languages.
• Build a stable experimentation platform with trackable results.
• Design and run experiments; document findings and push to production.
• Collaborate with founders in a highly technical environment with ownership of the roadmap.
• Train and run inference on Reson8's GPU cluster in NL datacenters.

Skills

Machine learning

Python

PyTorch

Experimentation infra

Research

Tools

NeMo

TensorRT

Kubernetes

Job description

About Reson8

At Reson8, we're on a mission to become a global leader in accurate, customisable speech-to-text for European languages, so every conversation is captured as it was meant to be.

Unlike generic tools, Reson8 understands domain-specific jargon and adapts to the unique needs of each professional environment. From medical consultations to legal depositions to restaurant operations, Reson8 handles the jargon that generic tools get wrong.

Available in 24 European languages so far, Reson8 is backed by world-class VCs and is committed to building a world where everyone is understood—no matter their language, accent, or jargon.

What You'll Build

• Train, fine-tune, and improve Reson8's proprietary speech-to-text models across 24+ European languages. You'll work hands-on with our model architectures to push accuracy on domain-specific vocabulary (medical, legal, financial) beyond what generic alternatives can do.

• Build and maintain a stable, reproducible experimentation platform—so the team can run training experiments with confidence, track results systematically, and iterate fast without breaking things.

• Design and run experiments: formulate hypotheses around model architecture, training recipes, data mixes, and domain adaptation strategies. Analyse results rigorously, document findings, and ship what works into production.

• Work directly with the founders in a highly technical environment: all three have engineering backgrounds from IMC and Adyen. You'll be in the room where decisions happen and have real ownership over our model roadmap.

• Train and run inference on Reson8's own GPU cluster in NL datacenters—owning the full loop from experiment to deployment instead of relying on opaque cloud compute.

• Stay close to the frontier: evaluate new architectures, training techniques, and tooling (NeMo, vLLM, TensorRT) and bring what works into our stack.

• Own your work end-to-end: identify research opportunities, run the experiments, build the infrastructure to support them, and validate results in production.

What You Bring

• Strong background in machine learning, with hands-on experience training and evaluating deep learning models (ideally in speech, NLP, or sequence-to-sequence domains).

• Proficiency in Python and PyTorch (strongly preferred). You've actually trained models, not just fine-tuned notebooks.

• Experience building or contributing to experimentation infrastructure—experiment tracking, reproducible training pipelines, evaluation frameworks, or similar tooling.

• Comfort reading research papers and translating ideas into working code.

• Bonus experience with:

  • Speech recognition architectures (CTC, RNN-T, encoder-decoder ASR, Whisper-style models).

  • NVIDIA tooling: NeMo, TensorRT.

  • Distributed training or multi-GPU workflows.

  • Rust (we use it in performance-critical parts of our stack).

  • Kubernetes or container orchestration.

  • Language-specific challenges in European languages (morphology, code-switching, diacritics).

How We Work

• Small, highly technical team where everyone is a shareholder and behaves like an owner.

• No passengers: expectation of high accountability, curiosity, and bias to action.

• High bar for craftsmanship and attention to detail that competitors ignore.

What We Offer

• Salary competitive with the top of the market

• Meaningful equity as an early employee with real influence over direction and culture.

• 25 days of vacation.

We're backed by Europe's best investors, and we're here to prove that fundamental AI infrastructure can be built on this continent, not just consumed from the US. If that matters to you, let's talk.

Amsterdam-based. Office-first, at least 3 days per week.