Вакансии

Задача — обучение с нуля специализированных моделей на 1B–10B параметров. – Подготавливать данные и выстраивать пайплайны обучения.

• Own AI features from concept to production in a high-load environment. • Build and fine-tune LLMs, train specialized models, and integrate them into the platform.

Kled AI is looking for a Research Scientist Benchmarks and Evals to develop benchmarks, evaluate frontier models, and demonstrate the value of opt-in human data.

Разрабатывать и исследовать ML-модели для генерации изображений и визуализации интерьерных/строительных решений на основе товаров Лемана ПРО. Проектировать и реализовывать end-to-end пайплайны для PoC: от сбора и подготовки данных до обучения, оценки и демонстрации результатов.

— Разрабатывать и поддерживать реализованные рекомендательные решения. — Искать новые гипотезы для улучшения пайплайнов обучения моделей и добавления новых признаков.

Location: Москва В подразделении риск-квантов (risk-quants) Сбера открыты вакансии для стажеров и специалистов с опытом работы, занимающихся моделированием финансовых рынков и машинным обучением (ML). Обязанности * заниматься разработкой моделей ценообразования и расчета риск-метрик для финансовых инструментов, а также подходов к управлению рыночными рисками * развивать передовые методы ИИ для финансов: генеративное моделирование мультимодальных данных (диффузия, VAE, LLM и др.), обучение с подкреплением * сочетать фундаментальные исследования с прикладными задачами Сбера. Требования * образование с математическим уклоном: предпочтительно (но не ограничиваясь) МКН СПбГУ, матфак ВШЭ, мех-мат МГУ, РЭШ, МФТИ, ИТМО и др. * понимание финансовой математики: знания в области теории вероятностей и случайных процессов, математической статистики и эконометрики * программирование: свободное владение Python (NumPy, Pandas, PyTorch) * базовые знание ML/DL * классические методы ML, Deep Learning * современные генеративные модели / RL-алгоритмы * архитектуры и принципы LLM. Будет плюсом: * наличие профильных сертификатов – CFA, FRM, CQF и т.д. * окончание курсов ШАД, Vega или аналогов * опыт работы над проектами в области GenAI, анализа временных рядов (Time Series, TS), обучения с подкреплением (RL). Условия * комфортный современный офис рядом с м. Кутузовская * формат работы: гибрид * ежегодный пересмотр зарплаты и годовая премия * корпоративный спортзал и зоны отдыха * более 400 образовательных программ СберУниверситета для профессионального и карьерного развития * расширенный ДМС, льготное страхование для семьи и корпоративная пенсионная программа * гибкий дисконт по ипотечному кредиту, равный 1/3 ключевой ставки ЦБ * бесплатная подписка СберПрайм+, скидки на продукты компаний-партнеров * вознаграждение за рекомендацию друзей в команду Сбера.

Location: Paris; Warsaw; London ABOUT MISTRAL Mistral provides full-stack AI solutions: from frontier models to developer tools, applications, and compute. We partner with enterprises tackling the hardest problems across high-stakes industries like finance, manufacturing, defense, healthcare, and the public sector, co-creating customized AI systems that they can run on their terms. We are a dynamic, collaborative team passionate about AI and its potential to transform society. Our diverse workforce thrives in competitive environments and is committed to driving innovation. Our teams are distributed between Europe, North America, Asia and the Middle East. We are creative, low-ego and team-spirited. As a Research Engineer on the Orchestral team, you will help build the agentic platform we use to train and evaluate our models. Orchestral is the system our researchers rely on to run training and evaluation at scale — from the rollout engine that powers agentic workflows to the infrastructure that keeps their workloads running reliably. What you build becomes the foundation the whole science team works on. Your work will span the platform's core: the library, tooling, abstractions and rollout engine — including new capabilities like long-running tasks, subagents and RL features — and the machinery that runs data and jobs at scale, with a focus on resiliency, routing and request handling. You will own significant pieces end to end, from design to production. This is a software engineering role rather than an ML or infrastructure-operations one: RL or PyTorch experience is not required, and you will spend your time building the platform, not operating it. You will be building a distributed RL system, with the scientists as your users. WHAT YOU WILL DO - Design and build the core of the Orchestral platform: the library, abstractions and tooling researchers use to train and evaluate models. - Build and evolve the rollout engine, shipping new capabilities such as long-running tasks and subagents. - Run Orchestral as a production system: keep training and evaluation workloads reliable at scale. - Improve how data and jobs flow through the platform as usage grows — scheduling, routing and request handling. - Treat researchers as your users: understand their workflows and turn recurring needs into robust, shared tooling. - Raise the bar on engineering quality across the platform: testing, code review, CI/CD, observability and documentation. - Contribute to open-source rollout-engine and agent-infrastructure projects where it benefits the platform. WHAT WE'RE LOOKING FOR - Experience directly building and debugging complex distributed systems at scale, with an emphasis on reliability and scalability. - Strong Python skills and solid software-design instincts: testing, code review, CI/CD. - A builder mindset: you have constructed systems from the ground up, not just operated polished ones. - A track record in small teams running complex systems at scale — for example in finance, crypto or ad tech. - Self-starter, low-ego and collaborative, with a product mindset toward your users. - Curiosity about ML and RL — this experience is not required, but a desire to learn is. NICE-TO-HAVES - Hands-on experience with Ray or similar distributed-computing frameworks. - Open-source contributions, especially to rollout-engine or agent-infrastructure projects such as SLIME, verl or PrimeRL. - Experience building or working with agentic workflows and environments. WHAT WE OFFER We offer a comprehensive benefits package designed to support your well-being, growth, and work-life balance. Benefits vary by country and may include healthcare coverage, parental leave, retirement plans, relocation support, wellness programs, meal and transportation allowances, and other location-specific perks. For the most up-to-date details on benefits available in your location, please refer to our Benefits page https://app.notion.com/p/mistralai/Benefits-at-Mistral-36e6ba59a7fe836b93dd01737fcc27ef?source=copy_link. PRIVACY POLICY Your privacy matters to us. You can learn more about how we handle your personal data in our Applicant Privacy Policy https://legal.mistral.ai/terms/applicant-privacy-policy.

- Сильный опыт в ML, Deep Learning, NLP и LLM. - Практический опыт создания AI-продуктов, работающих в production и в большом масштабе.

Location: Москва Команда управления развития искусственного интеллекта (блок «Развитие клиентского опыта B2C») ищет NLP/LLM-инженера для развития AI-copilot премиальных клиентских менеджеров. Мы переходим от классических NLP-решений к адаптивным мультиагентным системам, способным самостоятельно решать задачи в контуре бизнеса: поиск по базе знаний, суммаризация профиля, извлечение инсайтов из звонков, управление графиком, проактивные подсказки на звонке. Ищем Senior-инженера, который возьмет на себя архитектурную ответственность за решения с LLM под капотом, выстроит процессы оценки качества и будет писать чистый, поддерживаемый код, готовый к интеграции в продакшн выделенной бэкенд-командой. Обязанности * архитектура и разработка: проектировать и реализовывать end-to-end LLM-пайплайны, включая продвинутые RAG-системы и автономных AI-агентов с использованием планирования, function calling и мультиагентной оркестрации * дообучение и адаптация: SFT/RLHF/DPO, LoRA/QLoRA под специфичные бизнес-задачи. Принимать решения о том, когда дообучение необходимо, а когда достаточно промпт-инжиниринга или RAG * система оценки (Eval): поддерживать культуру качества, внедряя продуктовые и ML-метрики (RAGAS, DeepEval, кастомные LLM-as-a-Judge), автоматизировать регрессионное тестирование, связывать технические метрики с бизнес метриками * качество кода и интеграция: писать чистый, хорошо документированный и тестируемый python-код, который легко интегрируется и поддерживается бэкенд командой. Формировать стандарты разработки в рамках DS экспериментов и проектировать API-контракты для ML-компонент * менторство и влияние: участвовать в код-ревью, формировать технические стандарты команды, декомпозировать бизнес-задачи и предлагать оптимальные ML-решения. Работать напрямую с продуктом и стейкхолдерами. Требования * опыт в DL/NLP от 4 лет включая обучение/дообучение моделей на PyTorch, из них не менее года практического опыта с LLM * практический опыт построения AI-агентов и LLM-пайплайнов с использованием LangChain/LangGraph * глубокое понимание архитектуры современных LLM, включая механизмы attention, MoE, RoPE, KV-cache, методы alignment (RLHF/DPO), принципы эффективного инференса * сильная инженерная культура: Python (asyncio, pydantic, fastapi), написание чистого тестируемого кода (pytest), опыт разработки асинхронных микросервисов. Понимание Docker/K8s, CI/CD для ML * оценка и мониторинг: уверенное владение инструментами LLMOps (Arize, LangSmith, W&B, Phoenix). Опыт создания кастомных eval-пайплайнов и понимания того, как метрики влияют на продукт * работа с данными: опыт предобработки сложных данных для подготовки датасетов, знание SQL * автономность и коммуникация: Умение переводить неопределённые бизнес-задачи в технический план, аргументированно отстаивать архитектурные решения, работать без микроменеджмента. будет плюсом: * опыт опромышливания AI-агентов и знание принципов LLMOps, включая CI/CD, оркестрацию и управление версиями промптов * опыт интеграции ML-решений в высоконагруженные системы * понимание безопасности LLM-приложений. Условия * комфортный современный офис рядом с м. Кутузовская * гибридный формат работы после испытательного срока * ежегодный пересмотр зарплаты, годовая премия * корпоративный спортзал и зоны отдыха более 400 образовательных программ СберУниверситета для профессионального и карьерного развития * программа адаптации и помощь руководителя на старте * расширенный ДМС, льготное страхование для семьи и корпоративная пенсионная программа * гибкий дисконт по ипотечному кредиту, равный 1/3 ключевой ставки ЦБ * подписка Прайм с возможностью совместного использования на трёх близких * вознаграждение за рекомендацию друзей в команду Сбера.

Задачи: — участвовать в полном цикле разработки AI-агентов и универсальных ассистентов: от идеи и проектирования до запуска в продакшн — проводить эксперименты, тестировать гипотезы и исследовать новые подходы для создания агентских решений — разрабатывать и внедрять LLM-пайплайны, чат-ботов и интеллектуальные системы — обучать и дообучать NLP-модели под конкретные задачи продукта — интегрировать разработанные модели и логику в бизнес-процессы — мониторить качество работы моделей в продакшне, находить узкие места и оперативно вносить улучшения. Требования: — опыт в NLP и понимание технологий: от классических алгоритмов до трансформеров и генеративных моделей (LLM) — практический опыт коммерческой разработки AI-агентов, чат-ботов и LLM-пайплайнов — опыт обучения глубоких нейронных сетей — уверенное владение PyTorch, NumPy, Pandas, а также фреймворками для работы с агентами: LangChain, LangGraph (или их аналогами) — уверенное владение Python, Git, SQL, Bash, Docker. Умение писать чистый, поддерживаемый и тестируемый код — навык перевода бизнес-требований в конкретные технические и алгоритмические решения. Будет плюсом: — понимание принципов микросервисной архитектуры и опыт проектирования REST API — опыт разработки асинхронных и многопоточных приложений (asyncio, aiohttp, multiprocessing и др.) — опыт взаимодействия с реляционными и нереляционными базами данных — практический опыт настройки и поддержки CI/CD-пайплайнов (Jenkins, GitLab CI, ArgoCD или аналоги) — опыт развёртывания и управления приложениями в Kubernetes. Условия: — комфортный современный офис рядом с м. Кутузовская — гибридный формат работы — ежегодный пересмотр зарплаты, квартальная и годовая премия — корпоративный спортзал и зоны отдыха — более 400 образовательных программ СберУниверситета для профессионального и карьерного развития — программа адаптации и помощь руководителя на старте — расширенный ДМС, льготное страхование для семьи и корпоративная пенсионная программа — гибкий дисконт по ипотечному кредиту, равный 1/3 ключевой ставки ЦБ — подписка Прайм с возможностью совместного использования на трёх близких — вознаграждение за рекомендацию друзей в команду Сбера.

Location: London Workplace: Hybrid ABOUT NIANTIC SPATIAL We are building for the >80% of economic activity that happens beyond the reach of screens. AI has learned from the text, images and video online, but it still lacks an accurate understanding of physical space. That’s one of the biggest unsolved problems in AI, and it’s the one we’re solving at Niantic Spatial. We build spatial AI models and services so people and machines can reason, learn and act safely in the physical world. We turn camera imagery, video, sensor data and archives into accurate, AI-readable representations of real places that robots can train in and machines can use to know exactly where they are. In physical AI, mistakes have real consequences. That’s why we build AI grounded in measured reality, not guesswork, so the people and machines relying on it can trust it to get things right. Our team built some of the world’s most widely used maps. Now we’re building something bigger: a living, measured record of the real world that machines can query, learn from and use to understand how places change over time. If you want to help AI understand the world we actually live in, come build with us. ABOUT THE ROLE We are looking for 3D Reconstruction Research Scientist to join our Research Team in London. Our team researches and builds the 3D reconstruction technology behind Niantic Spatial's platform. These pipelines turn casually captured or domain specific imagery into high-fidelity explorable 3D reconstructions. This entails anything from Structure-from-Motion through to 3D Gaussian Splatting, feed-forward models, and generative models. Our team shipped 360 splatting to consumers and enterprise customers, developed the splatting tech that runs on phones, shipped multiple POCs to enterprise customers, and have innovated on countless AR technologies. Our team has published at top venues (CVPR, ICCV, ECCV, SIGGRAPH); more often than not, our papers end up in production. We're hiring a researcher to work at both ends of that pipeline. You will implement and iterate on the state-of-the-art for both academic publishing and production, regularly working on prototypes with a short turnaround into customer-facing production. You will work alongside world-class researchers in computer vision, machine learning, graphics, and robotics to develop new approaches for large-scale spatial understanding, localization, reconstruction, scene understanding, and geospatial foundation models. WHAT YOU’LL DO - Conduct original research in computer vision, machine learning, and spatial AI. - Design, implement, and evaluate novel algorithms for: - 3D reconstruction - Neural scene representations - 3D Gaussian Splatting - Geometric deep learning - Structure-from-Motion (SfM) - Meshing - Feed-forward models - Generative models - Spatial foundation models - Collaborate with research scientists and engineers to transition promising ideas into future products. - Develop large-scale experiments and benchmarking frameworks. - Work directly with the realities of production with real capture conditions, real hardware constraints, and real customers. - Publish research findings at leading conferences such as CVPR, ICCV, ECCV, or SIGGRAPH. WHAT YOU’LL BRING - PhD in Computer Vision, Robotics, Machine Learning, Computer Science, Mathematics, Physics, or a closely related field. Exceptionally strong candidates with equivalent experience in industry would be considered. - A strong research track record in 3D computer vision: reconstruction, depth estimation, novel view synthesis, SfM, SLAM, or closely related areas. - Strong publication record at top-tier conferences such as CVPR, ICCV, ECCV, SIGGRAPH, NeurIPS, etc. - Strong programming skills in Python and experience in PyTorch. - Evidence that you can carry an idea from research into something real; that may be a shipped feature, a production pipeline, a widely used open-source release, or similar. - Ability to independently drive research projects from idea generation through experimentation and publication. NICE TO HAVE - Experience with classical reconstruction tooling (COLMAP or similar). - C++ or CUDA experience, and a feel for performance and on-device constraints. - Experience with traditional 3D techniques — meshing, texturing, or bringing modern geometry or generative representations into native 3D. COMPENSATION & BENEFITS Base salary: Competitive market rate. Compensation also includes an annual bonus, equity, and a comprehensive benefits package. LOCATION & WORK MODEL This role is based in our London office, with three days per week in office. INCLUSIVE APPLICATION We know the strongest candidates don’t always tick every box. If you’re excited about this role and believe you could do it well, we encourage you to apply even if your experience doesn’t match every qualification listed — you may be exactly who we’re looking for. EQUAL OPPORTUNITY Niantic Spatial is an equal opportunity employer. Individuals seeking employment at Niantic Spatial are considered without regard to race, color, ancestry, national origin, religion, creed, age, gender (including pregnancy, childbirth, breastfeeding or related medical conditions), marital status, physical or mental disability, medical condition, genetic information, military or veteran status, gender identity, gender expression, sexual orientation, or any other protected category under applicable laws. Niantic Spatial will also consider qualified applicants with criminal histories in accordance with applicable laws. Please contact your recruiter if you want to request an accommodation for the job application or interview process. CANDIDATE PRIVACY I understand that by submitting my job application, the information I provide as part of that application will be used in accordance with Niantic Spatial’s Privacy Notice for Job Applicants and Candidates https://www.nianticspatial.com/applicant-privacy-notice.

Location: San Francisco, CA, USA About the role Chime's Data Science & Machine Learning team builds the models, services, and platforms behind how millions of members manage and grow their financial lives. We're hiring a Staff AI/ML Engineer on our Lending and Cashflow team to lead the modeling behind cash flow underwriting and surface member insight: using members' transaction and deposit data to make fair, accurate credit decisions for people that traditional credit scores often miss. In this role, you'll build and deploy machine learning systems on some of the richest transactional and behavioral data in fintech, turning it into the credit decisions behind products like MyPay, Instant Loans, and SpotMe. This is a highly applied role: you'll own problems end to end, from framing the question through to a model running in production and moving a metric that matters, and you'll set the technical direction for how Chime underwrites with cash flow data. In this role, you can expect to Lead the design, training, and deployment of deep learning and classical ML underwriting models that turn members' transaction and cash flow data into credit decisions, from approval and line assignment to loss forecasting Take models from problem framing through to production — training, evaluation, deployment, monitoring, and iteration — and stay accountable for how they behave once they're live Design and improve the systems around the model: feature pipelines, batch and real-time inference, monitoring, and retraining Partner with Product, Engineering, Analytics, and Risk to turn ambiguous business problems into ML solutions, and to make sure the solution is the right one Connect model performance to member outcomes and business metrics, and use experimentation to prove impact Set the technical roadmap for Lending ML and raise the bar across the team through design reviews, mentorship, and shared standards for model development, validation, and governance Help identify where AI/ML creates measurable impact for members — and where a simpler answer is the better one To thrive in this role, you have A track record of leading the design and deployment of deep learning models in production, including multi-quarter initiatives that span teams, with a solid grasp of architecture choice, training dynamics, and evaluation, and the judgment to make model design choices Solid machine learning fundamentals: classical modeling, evaluation design, and knowing which metric actually answers the question in front of you Hands-on experience across the end-to-end ML lifecycle — training, experimentation, optimization, deployment, and monitoring Comfort with messy real-world data, including label definition, leakage, class imbalance, and train/serve skew Strong proficiency in Python and SQL, with deep learning frameworks such as PyTorch and distributed compute such as Spark or PySpark Working knowledge of modern ML infrastructure — AWS and tools such as SageMaker, Airflow, Kafka, Redis, and Snowflake — and an MLOps mindset for keeping production systems healthy The ability to operate independently in ambiguous environments, influence technical direction beyond your own team, and explain model decisions clearly to engineers, risk partners, and senior leaders Nice-to-have Experience in either of these is a plus: Credit or underwriting models, and familiarity with model governance, explainability, and fairness requirements, such as adverse action reason codes and fair lending testing Cash flow or alternative-data underwriting: building credit features from bank transaction data, such as income and expense detection, balance trends, and transaction categorization Deployed transformer models in production #LI-Onsite #LI-DA1 Below is the base salary offered for this role and level of experience. Full-time employees may also be eligible for bonus(es), competitive equity, and benefits. For commissioned roles, the base salary listed in this job description does not include incentive/variable pay. The actual base salary offered may be higher, depending on your location, skills, qualifications, and experience. Wage Notice $246,000 — $340,000 USD A little about us At Chime, we believe that everyone can achieve financial progress. We created Chime—a financial technology company, not a bank*—on the premise that core banking services should be helpful, easy, and free. Through our user-friendly tools and intuitive platforms, we empower our members to take control of their finances and work towards their goals. Whether it's starting a savings account, purchasing a first car or home, launching a business, or pursuing higher education, we're proud to have helped millions unlock their financial potential. We're a team of problem solvers, dreamers, and builders with one shared obsession: our members. From day one, Chimers have worked tirelessly to out-hustle and out-execute competitors to bring our mission to life. Their grit and determination inspire us to work harder every day to deliver the very best experience possible. We each bring an owner's mindset to our work, refusing to be outdone and holding ourselves accountable to meet and exceed the highest bars for our teams, our company, and our members. We believe in being bold, dreaming big, and taking risks, while also working together, embracing our diverse perspectives, and giving each other honest feedback. Our culture remains deeply entrepreneurial, encouraging every Chimer to see themselves as stewards of our mission to help everyday Americans unlock their financial progress. We know that to achieve our mission, we must earn and keep people's trust—so we hold ourselves to the highest standards of integrity in everything we do. These aren't just words on a wall—our values are embedded in every aspect of our business, serving as a north star that guides us as we work to help millions achieve their financial potential. Because if we don't—who will? *Chime is a financial technology company, not a bank. Banking services provided by The Bancorp Bank, N.A. or Stride Bank, N.A., Members FDIC. What we offer for our full-time, regular employees 🏢 Our in-office work policy is designed to keep you connected - with four days a week in the office and Fridays from home for those near one of our offices, plus team and company-wide events depending on location. 💻 Benefits that support your work and life, including backup child, elder, and pet care and subsidized commuter benefits for eligible employees. ✨ Comprehensive health, financial, and wellbeing benefits designed to support you at every stage of life. 🏝 Generous vacation policy and company-wide paid days off 🫂 1% of your time off to support local community organizations of your choice 👟 Annual wellness stipend to use towards eligible wellness related expenses 👶 Up to 22 weeks of paid parental leave for birthing parents and 12 weeks of paid parental leave for non-birthing parents 👪 Access to family planning reimbursement 💚 A challenging and fulfilling opportunity to join one of the most experienced teams in FinTech and help millions unlock financial progress We know that great work can’t be done without a diverse team and inclusive environment. That’s why we specifically look for individuals of varying strengths, skills, backgrounds, and ideas to join our team. We believe this gives us a competitive advantage to better serve our members and helps us all grow as Chimers and individuals. Chime is proud to be an Equal Opportunity Employer. We consider qualified applicants without regard to race, color, ancestry, religion, sex, national origin, sexual orientation, gender identity, age, marital or family status, disability, genetic information, veteran status, or any other legally protected basis under provincial, federal, state, and local laws, regulations, or ordinances. We will also consider qualified applicants with criminal histories in a manner consistent with the requirements of state and local laws, including the San Francisco Fair Chance Ordinance, Cook County Ordinance, NYC Fair Chance Act, and the LA City Fair Chance Ordinance, and consistent with Canadian provincial and federal laws. If you have a disability or special need that requires accommodation during any stage of the application process, please contact: accommodations@chime.com . To learn more about how Chime collects and uses your personal information during the application process, please see the Chime Applicant Privacy Notice .

Location: 🇧🇷 São Paulo / 🌎Remote Workplace: Remote A little bit about us: We are Belvo, an open finance API platform with the bold vision of democratizing access to financial services in Latin America. We enable any financial innovator to access and interpret financial data, as well as initiate payments from their end-users accounts. We’re turning the messy complexities of the Latin American financial ecosystem into a modern set of tools to access and interpret data and move money in a seamless and secure way. We’re a highly-technical, passionate, and driven team. We are more than 90 people and our team currently represents 20 nationalities. We have offices in São Paulo and Mexico City – while a large portion of us work remotely. We are tackling a very stimulating problem: connecting fintech innovators with legacy financial infrastructure. We strive to go beyond the limits of what is possible today and to do so in an elegant and developer-first way. Since starting our adventure in May 2019, we have raised $71m from the leading VC investors globally. You can read more about our company here https://belvo.com/about/ and about our team and culture here https://belvo.com/careers/. Also, head to our blog https://belvo.com/blog/ for more news about what we’re building and how we work https://belvo.com/blog/category/working-belvo/. About the team: - We work in cross-functional, autonomous teams. We follow continuous delivery best practices executed on top of a modern technology stack.  - Our products are built for developers, by developers. Technological excellence is at the heart of what we do.   - We are pragmatic and customer-focused. We strive to find the right set of trade-offs in order to validate our hypothesis as early as possible, iterating on our products based on customer feedback.  - We communicate transparently. We do weekly all-hands where we get together to discuss company performance and goals. - While we are global and remote-friendly, we also operate from our vibrant offices in CDMX and Sao Paulo. To accommodate the various time zones in which we are based, we ensure we’re always synced up for at least 3 hours per day. - Also, we are backed by some of the leading investors in Silicon Valley and Latin America, including Founders Fund, Kaszek Ventures, and YCombinator. The challenge The Intelligence Squad is Belvo's AI enabler team. One part of the squad builds the platforms and patterns that let every team at Belvo ship AI. This role sits in the other part: turning the financial data and payments we aggregate across Latin America into models that make real decisions, who looks creditworthy, which transaction is fraudulent, what category a payment belongs to, whether an income statement checks out. That data drives real financial outcomes for the people and businesses on the other end, so the bar for rigor is high. You'll be the data science reference in the squad. We're not hiring someone to execute a modeling backlog someone else wrote. We want someone who decides which problems are worth modeling, defines how we know a model is good enough, and raises the data science bar for everyone they work with. You'll be deep enough in statistics and ML to defend a modeling choice to anyone, whether this is a set of rules to simplify a problem or a complex model that you build. You'll be close enough to the business to turn a commercial problem into a modeling roadmap and a business case. And you'll be a strong enough engineer to take your models from notebook to production and own them there. Responsibilities - Model development and selection: Lead the design of statistical and ML models (classification, regression, ranking, NLP or LLM-based) for problems like risk scoring, fraud detection and transaction categorization. Choose approaches on their merits and defend those choices with data, not habit. - AI-leveraged delivery: Use agentic coding and research tooling as a core part of how you work: exploring data, generating and testing feature hypotheses, scaffolding pipelines, writing the service around the model. We care about the decisions you ship, not hours at the keyboard, and we expect you to keep compressing the gap between the two. - Feature engineering with business context: Work with Product and the business teams to figure out which signals in Belvo's financial data actually explain the outcome you're modeling, and translate an ambiguous business question into a well-scoped modeling problem. - Productionization and ownership: Take models past the notebook. Build or shape the pipelines, APIs and services that run them reliably in production, and own them once they're live: monitoring, drift, retraining, and retirement when the evidence says so. - Experimentation: Design and run experiments (A/B tests and offline evaluation) to confirm that a model change actually moves the metric it's meant to move. - Cross-functional partnership: Work closely with Engineering, Data Research, Product, and the rest of the Intelligence Squad to ship end to end, and with stakeholders across the business to keep your models grounded in what actually matters to them. - Business and customer partnership: Work directly with stakeholders and, when needed, customers. Translate ambiguous business questions into well-scoped problems, and explain trade-offs and impact clearly to non-technical audiences, executives included. - Staying current: GenAI moves weekly. We expect you to be genuinely plugged in: model releases, papers, and tooling, to hold an opinion on what's real versus hype, and to bring the useful parts back to the squad as a demo, a doc, or a session for our company-wide AI Chapter. - Framing the problem: Decide what's worth modeling. Push back when a rule, a heuristic, or a better data pipeline solves the problem without a model, and be willing to kill or retire your own model when the evidence says so. This position may be for you if - 5+ years of experience as a Data Scientist, Applied Scientist, or in a similar hands-on modeling role. - You've owned models end to end in production, from framing the problem to monitoring it live, and you can point to the measurable business impact they had. - You have a solid grounding in statistics and can explain how your models work at that level, not just how to call them. - You've built and compared more than one type of model (for example tree-based methods, regression, deep learning) for the same problem and can justify why you picked the one you shipped. - You're comfortable in Python for data science (pandas, scikit-learn, and at least one of TensorFlow or PyTorch) and in SQL for pulling and shaping your own data. - You've taken several models you built into production yourself, not just handed it off to someone else to deploy.You http://deploy.You understand engineering fundamentals: testing, version control, CI/CD, and how to write code someone other than you can maintain. - You've built with LLMs past the API-call level: tool use or agents, retrieval, structured outputs, and shipped at least one to production. You can explain how you knew it was good enough, which means you've built some form of evaluation, not just eyeballed outputs. - Agentic coding tools are part of your daily workflow, not a novelty. You can walk us through a specific task that used to take you days and now takes hours: how you set it up, what you delegated, and where you still don't trust the output. - You're proficient in written and spoken English. Amazing if - You've worked in fintech, financial risk, or another regulated data environment before. - You have a business or strategy background alongside data science (an MBA, consulting, or a product/strategy role). - You've worked in a customer-facing data science role, such as a forward-deployed engineer or a solutions/applied team at a data platform company. - You've worked with MLOps tooling for model versioning, monitoring, and automated retraining. - You've deployed models on AWS (for example SageMaker or an equivalent managed ML service). - You've written evals or an LLM-as-judge harness for a non-deterministic system, and have opinions about where that approach breaks. - You've used LLMs inside the data pipeline itself: labeling, extraction, categorization, and can say where they beat a classical model and where they quietly don't. Tech stack - Python is our primary language for data science and modeling (pandas, scikit-learn, and deep learning frameworks as needed). - SQL against a cloud data warehouse for data access and feature work.Claude deployed company-wide and Claude Code for every engineer. Agentic tooling is part of how we work daily, not a side experiment. - Cloud infrastructure on AWS, favoring managed services so you can focus on the modeling problem rather than infrastructure plumbing. - An experimentation platform for A/B testing and offline evaluation. - We use Datadog for service observability across the company. - We follow Continuous Integration and Continuous Delivery best practices. Our tech stack: - We’re building our platform using modern technologies, putting the focus on reliability and long-term maintainability - We primarily use Python on the backend. We use battle-tested technologies such as Django, and we are heavy users of Python’s asyncio for some parts of our stack - We use Javascript, Vue.js and Sass on the frontend. We are developing and leveraging our own design system and component library - We run our infrastructure on top of Amazon Web Services, leveraging managed services. We tend to favor the use of managed services wherever possible, in order to focus on our business problems - We observe and monitor our services using Datadog - We follow Continuous Integration and Continuous Delivery best practices Our process steps: At Belvo every hire is so important to us, that we share the decision to hire as a team. - People team chat - Take-home challenge - Challenge presentation - Team fit chat - Meet the founders Our perks: 🚀 Stock options (we are all owners and this is very important to us) 📈 Annual company bonus linked to company performance 🔆 Flexible working hours 🔝 Remote friendly 🐶 Pet friendly 👩🏻‍⚕️ Health Insurance 🎊 Paid time off on your birthday 💻 Renew your laptop every 3 years 🙋 Training Budget 😎 Team building events 🙀 Bank holidays swap within 30 days 🔋 Fitness/ wellness stipends 🚀 Yearly company offsite 🍒 Fresh fruit every week, all-you-can-drink tea and coffee 🏝 Extra days off accrued when completing 2, 3, 4 years and beyond at Belvo 🎳 Yearly department offsite

• Взаимодействие с data scientists, инженерами, аналитиками и бизнес-стейкхолдерами. • Коммерческий опыт в ML/ML Engineering от 3 лет, уровень Middle+ / Senior.

Location: Москва Мы — технологическая команда, создающая инновационные решения для спорта. Наш продукт направлен на трансформацию и объективизацию судейства. Мы ищем опытного специалиста в области MLE для участия в разработке и развитии наших решений. **Предстоит** работать с ключевым функциональным и архитектурным ядром нашей системы видеоаналитики, а именно: Обязанности * работать со скелетом как с многомерным временным рядом: сегментация выступления на элементы и фазы, детекция границ (отрыв, приземление, точки экстремума), выделение повторяющихся циклов, temporal action detection на видеопотоке, сглаживание и агрегация предсказаний по времени * участвовать в проектировании API, внутренних контрактов и структуры данных * оптимизировать модели для достижения целевых показателей задержки * развивать высокопроизводительный Python-фреймворк для CV/ML-пайплайна на более чем 15 моделей, работающих единовременно * реализовывать механизмы обработки с multiprocessing и multithreading. Обеспечивать эффективную оркестрацию между CPU и GPU-ресурсами * минимизировать накладные расходы на межпроцессное взаимодействие, копирование данных и работу с памятью. Проводить профилирование и устранять узкие места в video I/O, decode, inference и postprocessing * проводить профилирование и устранять узкие места на всех этапах: Video I/O, декодирование, препроцессинг, инференс, постобработка. Требования * опыт работы ML Engineer / CV Engineer от 3 лет * высокое качество кода на Python (OOP, Pydantic, SQLAlchemy, для DS - NumPy, OpenCV, PyTorch) * опыт реализации сложной вычислительной или бизнес-логики * практический опыт разработки production-ready ML/CV pipeline * опыт работы с видеоаналитикой, обработкой видео или потоковых данных * понимание особенностей Python runtime для высокопроизводительных пайплайнов: GIL, IPC, serialization, shared memory * опыт использования инструментов ускорения инференса: ONNX, TensorRT, TorchScript или аналогов * опыт работы с Linux, Docker, работы в команде в Git, CI/CD Будет плюсом: * опыт разработки расчетных систем * PostgreSQL, ClickHouse, SQLAlchemy * владение компилируемыми языками, такими как Go или C++, для реализации или оптимизации критичных по производительности частей системы * опыт работы со спортивной аналитикой или системами судейства. Условия * комфортный современный офис рядом с м. Кутузовская  * гибридный формат работы * ежегодный пересмотр зарплаты, квартальная и годовая премия * корпоративный спортзал и зоны отдыха более 400 образовательных программ СберУниверситета для профессионального и карьерного развития  * программа адаптации и помощь руководителя на старте  * расширенный ДМС, льготное страхование для семьи и корпоративная пенсионная программа  * гибкий дисконт по ипотечному кредиту, равный 1/3 ключевой ставки ЦБ * подписка Прайм с возможностью совместного использования на трёх близких  * вознаграждение за рекомендацию друзей в команду Сбера.

Location: Austria; France; Germany; Italy; Netherlands; Poland; Spain; United Kingdom Why Kimchi? Great models are raw ingredients. Kimchi is what happens when you ferment them with good engineering. We started inside CAST AI helping companies run LLMs on their own Kubernetes clusters; today we run our own coding harness on any model you pick, and the router deciding which model answers each request is ours too. The Auto Router Our promise is one enforcement layer for your AI: budgets, spend attribution, and model routing, applied before a request ever hits a model. The Auto Router makes the call: for every request, which model serves it. Eight open-source models whose input prices span more than 20×, cached input up to 50× cheaper than fresh, frontier providers on customers' own keys, and a coding agent that fires hundreds of requests before its task is done. Frontier-level task success at a fraction of the token cost: that sentence is the product, and it stays true only if the model inside the router keeps getting better. That model is yours. Made by Cast AI — a Miami-headquartered, globally distributed team across 34 countries that hit unicorn status in January 2026 with a $50B+ backed investment pushing its valuation past $1 billion. About the role Every request gets a decision: which model serves it. This is an applied-ML seat, not an infrastructure one. Platform engineers own the proxy, the streaming path, and the fleet. You own the model inside them: the training data, the objective, the evaluation, and the iteration loop that ships a better router every cycle. The problem: routing is a prediction problem wearing a product's clothes. For every request you predict the quality each candidate model would deliver, what it would cost, and how fast it would decode, and you have to be right enough, in aggregate, that the savings are real and the quality never dips below frontier. The data is messy: rollout outcomes from benchmark harnesses, per-request records from the live fleet, override signals, and token streams whose speed depends on the serving concurrency at the moment. Speed itself is a correctness variable: on deadline-bound agentic work, a model that decodes three times slower doesn't do the same work slower, it does less work. That's why naive cost-per-token objectives quietly optimize for the wrong thing. And the bar for evidence is high: a router change ships only after production-parity evaluation with error bars, then gets held to its numbers in live traffic. Stack: Python; PyTorch; SQL at scale (ClickHouse); OpenAI- and Anthropic-compatible LLM APIs; benchmark and rollout harnesses (RouterBench-class); GitLab CI. Requirements You've spent 5+ years in applied ML or ML research, and you've trained, evaluated, and shipped models that make real-time decisions (ranking, recommendation, bandits, pricing, or LLM routing) and you can tell that story with receipts. Strong Python and SQL at scale: you can pull, clean, and interrogate millions of request-level records without waiting on anyone. You know the difference between a model that got better and an eval that got easier. You have real opinions about what LLM benchmarks measure, agentic harnesses included. You know enough serving reality (TTFT, decode speed, batching, prefix caching, concurrency) to model the objective honestly. You instrument before you optimize, error bars and all. A wide mandate excites you more than it unsettles you. Nice to haves You've worked on LLM routing specifically: RouterBench-class evals, cost/quality/latency Pareto fronts. Calibration, abstention, or conformal methods in production decision systems. You've run online A/B loops with drift monitoring on a model that touches real money. Responsibilities Own the routing model end to end: features, objective, training, calibration, and release cadence. Iterate it like a product, not a research artifact. Build the evaluation that gates every change: production-parity benchmarks, router-vs-router comparisons, holdouts by task family, error bars on pass rates. Nothing ships on vibes. Turn fleet data into training signal: per-request outcomes from production, benchmark rollouts, and override feedback, with failure categorization (infra vs task) so poisoned labels never reach the policy. Model the objective the user actually experiences: task success × token cost × wall-clock speed under the real concurrency regime, with cached-input economics in the picture. Run the online loop: shadow, A/B, full rollout, drift monitoring, retrain. When production disagrees with your offline eval, you find out why before the next ship. Push the frontier where it pays: calibrated decisions, abstention under uncertainty, speed-aware routing. We experiment on what a router should be, not just which model it picks today. Partner with platform engineers on instrumentation: the features and labels recorded per request (decode speed first-class among them) and the experiments the data path must support. What’s in it for you? Competitive salary (depending on the level of experience). Enjoy a flexible, remote-first global environment. Collaborate with a global team of cloud experts and innovators, passionate about pushing the boundaries of Kubernetes technology Equity options. Get quick feedback with a fast-paced workflow. Most feature projects are completed in 1 to 4 weeks. Spend 10% of your work time on personal projects or self-improvement. Learning budget for professional and personal development - including access to international conferences and courses that elevate your skills. Annual hackathon to spark new ideas and strengthen team bonds. Team-building budget and company events to connect with your colleagues. Equipment budget to ensure you have everything you need. Extra days off to help maintain a healthy work-life balance. Hiring process Screening call with Recruiter Hiring Manager interview Technical interview (system design) Live coding Culture Check interview with an executive As part of our standard hiring process, we would like to inform you that a background check may be conducted at the final stage of recruitment through our third-party provider, Checkr. Please note that Cast AI does not provide any form of visa sponsorship/work permit. #LI-Remote

Location: Austria; France; Germany; Italy; Netherlands; Poland; Spain; United Kingdom Why Kimchi? Great models are raw ingredients. Kimchi is what happens when you ferment them with good engineering. We started inside CAST AI helping companies run LLMs on their own Kubernetes clusters; today we run our own coding harness on any model you pick, and we run the models underneath it ourselves. Kimchi Inference One API line and a customer's existing client just works: DeepSeek, GLM, Kimi, MiniMax, Nemotron served serverless from our own GPU fleet (OpenAI- and Anthropic-compatible, pay per token, spend visible per user, team, and key). When control matters, the same models run on the customer's own cluster instead. Behind that one line sits the machinery you'll own: SGLang and vLLM engines, B300 nodes, a shared cluster filesystem that holds every checkpoint exactly once, KV caches riding coding traffic with heavy prefix reuse, and the benchmark harness that gates every change we ship. Made by Cast AI — a Miami-headquartered, globally distributed team across 34 countries that hit unicorn status in January 2026 with a $50B+ backed investment pushing its valuation past $1 billion. About the role Throughput. Latency. Cold starts. Cost per token. Move those four numbers in the right direction and two things happen: customers get faster, cheaper inference, and our margins improve. That's the entire thesis of this role. Every loader you tune, every scheduler decision you land, every kernel you stare at shows up in a customer's p99 and on our P&L. Your mandate is the inference stack's optimization, whatever it takes, and the technical direction of how we optimize it is yours to argue with data. The problem: the checkpoints run 0.3-1.5 TB, and the "right" serving configuration depends on traffic shape, sequence-length distribution, batch dynamics, GPU SKU, memory bandwidth, and quantization tolerance, all variables that shift week to week. Most teams pick a configuration once, over-provision GPUs, and absorb the cost. We don't: we took Kimi-K3 (1.5 TB) from over an hour cold-to-serving to under twelve minutes with a shared filesystem, O_DIRECT weight loaders, and node-local JIT caches, and we're nowhere near done. Prefill/decode disaggregation and tiered KV caches are next. Stack: Python; SGLang; vLLM; TensorRT-LLM; PyTorch; fastsafetensors; CUDA-adjacent tooling; Kubernetes; virtiofs shared filesystems; ClickHouse; OpenTelemetry; Prometheus; Grafana; GitLab CI. Requirements You've built real ML systems for 5+ years, and your portfolio shows depth in inference or training infrastructure, not just model training notebooks. Your Python is production-grade: services, not scripts. You've run vLLM, SGLang, or TensorRT-LLM for real, and you have a working mental model of why an engine performs the way it does on a given GPU. You've measured quantization regressions, not just compression ratios. Multi-GPU and multi-node setups hold no fear: tensor and expert parallelism, collective communication, and their practical failure modes. You know where the bytes live (page caches, O_DIRECT, network filesystems under terabyte-scale reads) and how fast they move. You instrument before you optimize, and you can tell a real win from a benchmark artifact. A wide mandate excites you more than it unsettles you. Nice to haves You've worked on LLM routing specifically: RouterBench-class evals, cost/quality/latency Pareto fronts. Calibration, abstention, or conformal methods in production decision systems. You've run online A/B loops with drift monitoring on a model that touches real money. Responsibilities Push throughput. Continuous batching, chunked prefill, speculative decoding with draft models, CUDA-graph capture, and scheduler tuning across the engine fleet. Find the ceiling on each GPU SKU, then raise it. Cut latency. Attack TTFT and inter-token latency separately; profile to find the actual bottleneck (compute, memory bandwidth, scheduling, storage, or network) and fix it, not the one you assumed. Get more out of the KV cache. Hierarchical and paged attention, prefix caching on heavy-reuse coding traffic, eviction policies, quantized KV. A lot of the unrealized throughput lives here, and you'll own it. Kill cold starts. Weight loaders (O_DIRECT, fastsafetensors-class), JIT and compile caches, shared-filesystem read paths. A 1.5 TB checkpoint has no business taking an hour to reach serving. Quantize without regressing quality. INT8/INT4, FP8/FP4 across weights, activations, and KV, measured on real workloads, not just perplexity benchmarks. Scale across nodes. Distributed inference topologies, network-aware placement, prefill/decode disaggregation, checkpointing strategies that don't bottleneck on storage or interconnect. Set the technical direction. Decide what we benchmark, what we adopt, and what we build ourselves. Bring the team along with strong writeups and reproducible experiments. What’s in it for you? Competitive salary (depending on the level of experience). Enjoy a flexible, remote-first global environment. Collaborate with a global team of cloud experts and innovators, passionate about pushing the boundaries of Kubernetes technology Equity options. Get quick feedback with a fast-paced workflow. Most feature projects are completed in 1 to 4 weeks. Spend 10% of your work time on personal projects or self-improvement. Learning budget for professional and personal development - including access to international conferences and courses that elevate your skills. Annual hackathon to spark new ideas and strengthen team bonds. Team-building budget and company events to connect with your colleagues. Equipment budget to ensure you have everything you need. Extra days off to help maintain a healthy work-life balance. Hiring process Screening call with Recruiter Hiring Manager interview Technical interview (system design) Live coding Culture Check interview with an executive As part of our standard hiring process, we would like to inform you that a background check may be conducted at the final stage of recruitment through our third-party provider, Checkr. Please note that Cast AI does not provide any form of visa sponsorship/work permit. #LI-Remote

Location: UK, London Workplace: OnSite Here at Humanoid, we believe in a future where robots amplify human potential. That’s why we’ve set out on a mission to build the world’s most capable, commercially-scalable, and safe humanoid robots. We’re bringing that mission to life with HMND‑01 - our rapidly developed humanoid platform being deployed in real industrial environments - and we’re growing the team to take it even further. ABOUT THE ROLE We are looking for a strong Deep Learning Engineer to join our Autonomy team and develop the learning-based algorithms that power dynamic locomotion and whole-body control on our humanoid robots. You will develop robust, reusable motion capabilities that enable the robot to move dynamically, maintain balance, interact with its environment, and combine locomotion with manipulation. This includes building the underlying learning-based motion primitives that can be composed into increasingly complex whole-body behaviors. You will work closely with robotics, controls, simulation, and machine learning engineers to turn learned behaviors into reliable capabilities running as part of the robot's real-time control stack.   WHAT YOU'LL DO - Design and train learning-based locomotion and whole-body control policies for dynamic robotic behaviors. - Develop motion primitives using different learning based generative models. - Apply reinforcement learning, imitation learning, and other learning-based approaches to motion generation and control. - Build scalable simulation, data generation, and training pipelines using platforms such as Isaac Lab and MuJoCo. - Develop methods for robustness, adaptation, and sim-to-real transfer, including domain randomization and learning from real-world robot data. - Deploy, evaluate, debug, and fine-tune learned policies on physical humanoid robots. - Optimize policy inference and training systems for the performance and latency requirements of real-time control. - Integrate learned policies into the broader robot control and autonomy stack.   WHAT WE'RE LOOKING FOR - BS, MS, or PhD in Robotics, Machine Learning, Computer Science, or a related technical field. - 3+ years of experience developing machine learning systems in industry or research. - Strong understanding of deep learning and modern policy architectures, including training, inference, experimentation, and debugging. - Hands-on experience with RL and at least one additional area such as imitation learning, generative models, sequence models, or large-scale representation learning. - Strong Python and PyTorch experience. - Experience with physics-based simulation environments such as Isaac Lab, Isaac Sim, MuJoCo, or similar platforms. - Ability to reason about learning algorithms in the context of dynamics, control, and physical systems. Nice to have: - Experience applying reinforcement learning or imitation learning to legged robots, humanoids, or other complex physical systems. - Experience with whole-body control, locomotion, motion imitation, or loco-manipulation. - Experience deploying learned policies on real robotic systems and debugging the sim-to-real gap. - Familiarity with motion capture, trajectory optimization, model-based control, or learning from demonstrations. - Experience training policies at scale across large numbers of simulated environments. - Publications at NeurIPS, ICML, ICLR, CoRL, RSS, CVPR, or equivalent contributions to impactful open-source robotics or machine learning projects.   WHAT WE OFFER - Competitive equity: stock options with meaningful upside as we scale. - 30+ paid days off, including 23 days of annual leave, all UK bank holidays, and additional company closure days (including Christmas–New Year shutdown). - Private healthcare, including virtual and in-person care. - Pension scheme with 8% total contribution (5% employee, 3% employer) on full earnings. - Free daily breakfast, catered lunch, and snacks in-office. - Work at the frontier - collaborate daily with world-class engineers, researchers, and product experts building the next generation of AI and humanoid robotics. - Real ownership - direct access to founding leadership, meaningful input on product direction, and the ability to drive key initiatives from day one.