Senior Research Scientist | Model Steering

DeepL·London·22 июля

Senior Research Scientist | Model Steering

🏢 ОфисSeniorПолная занятость

Зарплата не указана

Вилки нет, про деньги придётся договариваться с нуля.

Нажмите на сигнал, чтобы увидеть, на чём он основан

Наша компания

MEET DEEPL DeepL is a global AI product and research company focused on building secure, intelligent solutions to complex business problems. Over 200,000 business customers and millions of individuals across 228 global markets today trust DeepL's Language AI platform for human-like translation, improved writing and real-time voice translation.

О роли

Founded in 2017 by CEO Jaroslaw “Jarek” Kutylowski, DeepL now has around 1,000 passionate employees and is supported by world-renowned investors including Benchmark, IVP, and Index Ventures. Our goal is to become the global leader in trusted, intelligent AI technology, building products that drive better communication, foster connections, and create a meaningful impact. To achieve this, we need ta

Чем предстоит заниматься

We are looking for a Senior Research Scientist to lead fine-tuning, post-training, model-steerability, and reinforcement learning for the next generation of DeepL's LLM-based translation models. This is a high-impact, hands-on role for a researcher who can own a major research direction, prototype rapidly, run large-scale experiments, and drive breakthroughs all the way into production

Your central focus is how we blend data and shape model behaviour after pre-training. You will fuse high-value human expert data with synthetic data, and you will lead the effort to make our translation models steerable, following custom user instructions, rules, and context. You will work with model architectures at the scale of hundreds of billions of parameters

Drive the development of translation models that are steerable conditioned on user preferences, rules and context

Drive hands-on research and development on post-training for our core translation models: supervised fine-tuning, knowledge distillation, preference optimization, and reinforcement learning tuned to translation quality

Build reward models and evaluator models for translation, including rubric- and reference-based grading, and investigate and mitigate reward hacking and quality-estimation failure modes

Drive an agenda toward models that ingest multimodal content and context to increase translation quality

Own the full lifecycle of model delivery: prototyping, ablations, training, evaluation, optimization, and production deployment, working closely with engineering to ship into real-time systems at scale

Establish strong practices for evaluation, reproducibility, monitoring, and continuous model improvement in production

Mentor researchers and engineers, promote hands-on collaboration, and raise the bar for model quality

Наши требования

Demonstrated experience fine-tuning and training large models at scale, including distributed/multi-node training (e.g. FSDP, DeepSpeed, or Megatron-style frameworks) and efficient training techniques

Experience fine-tuning existing reasoning models for specific tasks and behaviors without degrading their reasoning capabilities

Experience with machine translation, multilingual NLP, or language quality estimation

Familiarity with inference and serving at scale (e.g. via vLLM, SGLang, TensorRT-LLM, etc) and long-context modelling

Publications at top-tier venues

Proven experience making large models steerable and instruction-following by identifying the most effective method to instill a given behavior, drawing from instruction tuning, latent space methods, steering vectors, and/or constrained encoding and decoding methods

Deep, hands-on expertise in LLM post-training (SFT, DPO), knowledge distillation (teacher-student training), and/or reinforcement learning (RLHF/RLAIF, PPO/GSPO, and reward modeling)

Strong data-centric instincts for building synthetic-data and preference-data pipelines, LLM-as-judge generation, data curation and filtering, and reasoning about data mixtures and ablations

Experience designing evaluation and reward signals using automatic metrics, LLM-as-judge evaluation, non-verifiable rewards, and human-in-the-loop evaluation

A hands-on builder who enjoys training models, running experiments, debugging pipelines, and integrating ML systems into production while staying grounded in product impact and real-world quality

Ownership of a substantial research direction with strong execution, and experience mentoring others on a fast-moving, applied research team

Strong coding and experimentation skills (Python, PyTorch/JAX/Tensorflow), and the ability to communicate clearly and align research with product and engineering priorities

Мы предлагаем

Diverse and internationally distributed team: joining our team means becoming part of a large, global community with people of more than 90 nationalities. We're more than just colleagues; we're a group of professionals with a shared mission to connect diverse cultures. Our global presence is growing–we've doubled in size nearly every year, with our employees based in the UK, Germany, the Netherlands, Poland, the US, and Japan, and we continue to expand our network

Open communication, regular feedback: as a language-focused company, we value the importance of clear, honest communication. We value smooth collaboration, direct and actionable feedback, and believe that leading with empathy and growth mindset makes us better together

Hybrid work, flexible hours: we offer a hybrid work schedule, with team members coming into the office twice a week. This allows you to engage directly with your team and experience the unique energy of our workspace, while still enjoying the flexibility and comfort of working from home. With flexible working hours and trust in your productivity, we are in sync with your team’s general locations and time zones to foster effective and seamless collaboration

Virtual Shares - An ownership mindset in every role. We believe everyone should share in our success, and that’s why every employee receives Virtual Shares, linking your contribution directly to DeepL’s growth and rewarding you with a stake in our future

Regular in-person team events: we bond over vibrant events that are as unique as our team, from local team and business unit gatherings, to new-joiner onboardings, to company-wide events that bring us all together–literally

Monthly full-day hacking sessions: every month, we have Hack Fridays, where you can spend your time diving into a project you're passionate about and get the opportunity to work with other teams–we value your initiatives, impact, and creativity

30 days of annual leave: we value your peace of mind. With 30 days off (excluding public holidays) and access to mental health resources, we make sure you're as strong mentally as you are professionally

Competitive benefits: just as our team spans the globe, so does our benefits package. We've crafted it to reflect the diversity of our team and tailored it to align with your unique location, to ensure you feel supported every step of the way

If this role and our mission resonate with you, but you're hesitant because you don't check all the boxes, don't let that hold you back. At DeepL, it's all about the value you bring and the growth we can foster together. Go ahead, apply—let's discover your potential together. We can't wait to meet you!

You are welcome at DeepL for who you are - we appreciate authenticity here. Our product is for everyone, and so is our workplace. The more voices we have represented and amplified in our business, the more we will all succeed, contribute, and think forward! So bring us your personal experience, your perspectives, and your background. It’s in our diversity that we will find the power to break down language barriers in the world

DeepL

London

ГрейдSenior

ЗанятостьПолная занятость

РегионВеликобритания

ФорматОфис

ИсточникСкрыто

Опубликовано22 июля

Все вакансии компании

AI-помощник

под эту вакансию

Войди, чтобы AI оценил твоё соответствие вакансии и написал сопроводительное письмо

Мы против мошенников на площадке: если тебя просят заплатить, продиктовать код или установить непонятное приложение, прекращай общение и сразу пиши нам (чат с основателем или форма обратной связи).