Вакансии

Задачи: — Повышение эффективности утилизации GPU: формирование гипотез, исследование способов повышения эффективности, участие в реализации и внедрении решений. — Формирование рекомендаций и лучших практик по повышению производительности GPU-инфраструктуры. — Оптимизация и профилирование: поиск узких мест в производительности и их устранение с помощью профилировщиков, оптимизация доступа к памяти, ядрам, времени ожидания и пропускной способности. — Развитие инструментов диагностики: создание и улучшение инструментов для выявления и устранения инфраструктурных проблем, влияющих на эффективность утилизации, стабильность и скорость GPU-вычислений для обучения и инференса. — Исследование и внедрение современных решений: изучение новейших подходов к организации инфраструктуры для обучения и инференса, оценка их эффективности и внедрение в проекты. — Анализ архитектуры, тестирование, интеграция: взаимодействие с разработчиками, ML-инженерами и системными архитекторами, участие в оценке аппаратных решений, предложение улучшений для будущих поколений GPU, разработка планов тестирования, формирование бенчмарков, анализ регрессий производительности. Требования: — Знание Python и опыт системного программирования, разработка библиотек или фреймворков. — Опыт работы с PyTorch и распределённым обучением через torch.distributed. — Применение подходов параллелизации (data parallelism, tensor parallelism, pipeline parallelism, expert parallelism) для распределённого инференса или обучения. — Интерес к LLM и MLOps, понимание задач и вызовов эксплуатации больших моделей в продакшене. — Опыт работы с GPU (NVIDIA) и CUDA: понимание архитектуры GPU, разработка или оптимизация алгоритмов под CUDA, использование Nsight, nvprof или аналогов. — Умение оптимизировать производительность GPU-приложений и повышать эффективность утилизации GPU. — Способность анализировать профили и метрики производительности. — Навыки чтения и оптимизации сложного кода. — Эффективная командная работа и готовность делиться знаниями. Будет плюсом: — Уверенное владение C/C++ или аналогичными низкоуровневыми языками. — Опыт работы с библиотеками RL-обучения для LLM: veRL, slime, NeMo-RL, SkyRL и другими. — Опыт работы с библиотеками инференса: vLLM, SGLang, TRTLLM. — Опыт оптимизации под реальные продакшен-нагрузки, работа с low-latency- или real-time-системами.

Location: Munich; Warsaw; London Workplace: Hybrid 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. THE ROLE The Inference Foundation team owns the core of Mistral's inference stack: the inference engine and its orchestration, from the feature set and configuration that serve our models in production to the release machinery that keeps the stack current and production-grade. This is a hybrid position spanning production LLM serving, engine and platform development, and capacity engineering. You will work on three intertwined problems: 1. Optimize the inference stack at scale — feature development and fixes in the engine and orchestrator, squeezing more throughput and lower latency out of every GPU under strict quality-of-service targets. 2. Make capacity elastic — scaling up and down should be cheap and fast, not a performance cliff. 3. Power the training of our frontier models — high-performance serving that keeps RL and post-training loops running at full speed.   What You Will Do INFERENCE ENGINE & ORCHESTRATION - Develop and fix the core of the inference stack — engine and orchestrator — including feature selection, configuration, and tuning for maximum performance at scale - Own the release process for the serving stack: validated, regression-free releases through automated performance gates and progressive rollout - Drive improvements and fixes upstream when the open-source engine is the right place for them PERFORMANCE & CAPACITY AT SCALE - Optimize serving efficiency across the fleet — driving down pod startup time, tackling cold-cache regressions on scale-up, smarter caching and offloading - Optimize and maintain the optimal serving topology — overlap communication and transfers with computation, ensure optimal placement, connectivity, and routing SERVING FOR FRONTIER TRAINING - Build the serving infrastructure that powers RL and post-training for our frontier models - Optimize inference performance across the full spectrum of our workloads   What We're Looking For - Experience building and running ML/LLM services at scale, with clear latency and availability targets - Hands-on experience with inference engines such as vLLM, SGLang, TensorRT-LLM, or others - A solid grasp of inference internals: prefill vs. decode, KV-cache behavior, batching, scheduling, speculative decoding, parallelism strategies - Familiarity with distributed and disaggregated serving architectures - Comfortable debugging across the full stack — CUDA/NCCL, kernels, containers, networking, storage - Python for systems tooling and backend services; PyTorch - Kubernetes for running infrastructure at scale - GPU and networking fundamentals: CUDA runtime, NCCL, InfiniBand/RDMA It Would Be Great If You Have - Demonstrated vLLM/sglang know-how — upstream contributions, or a track record of running in demanding production environments - Hardware-aware optimization for various model architectures - Experience serving MoE models at scale (expert parallelism, expert placement/load balancing) - CUDA/Triton kernel development; Nsight Systems/Compute profiling - Rust and/or C++ in production systems 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.

Задачи: — Разрабатывать, обучать и внедрять ML/DL-модели для бизнес-задач. — Строить AI-агентов и multi-agent workflows с tool/function calling, MCP и LangGraph/LangChain (или аналогами). — Проектировать RAG-системы: эмбеддинги, hybrid search, reranking и векторные БД. — Выполнять fine-tuning LLM, оптимизировать inference и интегрировать модели через API и микросервисы. — Участвовать в MLOps/LLMOps: evaluation, мониторинг, tracing, guardrails и повышение качества AI-систем. Требования: — Опыт: от 2 лет в ML/DL или разработке AI-систем. — Стек: Python, NumPy/pandas, scikit-learn, PyTorch / TensorFlow, Hugging Face Transformers. — Знания: понимание ML/DL, архитектуры трансформеров и принципов работы LLM. — Навыки: практический опыт с AI-агентами, RAG, vector DB и structured output; уверенный Git и деплой в production. — Языки: знание узбекского и английского (Intermediate+) языков. Будет преимуществом: — Разработка API на FastAPI / Django REST Framework и контейнеризация с Docker. — Computer Vision и видеоаналитика (OpenCV, YOLO, tracking, RTSP). — MLOps/LLMOps-инструменты, self-hosted inference, GPU и Kubernetes. — Мультимодальные модели или голосовые пайплайны (STT-LLM-TTS). Условия: — Развитие: оплачиваемые профильные курсы, IT-конференции, фестивали и тимбилдинги. — Условия: бесплатные вкусные обеды в офисе. — Карьера: реальные масштабные задачи, влияние на продукты компании, профессиональный и финансовый рост. — Локация: офис в г. Ташкент. — График: 5/2, с 09:00 до 18:00. — Командировки: в г. Коканд (организуются и оплачиваются компанией). — Зарплата: От 14.000.000 сум.

Задачи: — Ускорять и масштабировать LLM в продакшене: оптимизация инференса очень больших моделей (до 1T+ параметров), SGLang, KV/prefix caching, speculative decoding, квантизация. — Разворачивать распределенный инференс: MoE, tensor/expert/pipeline parallelism на multi-node / multi-GPU кластерах (32x H200 + AWS). — Заниматься Post-Training: обучать и дообучать модели компаньонов (SFT, DPO/RLHF), улучшать агентные каркасы и алгоритмы чата. — Осуществлять техлидство (без бюрократии и PM): ревью экспериментов команды NLP/CV (7 инженеров), выбор направлений и оценка гипотез. Требования: — Глубокий опыт оптимизации LLM в продакшене (SGLang, vLLM или TensorRT-LLM). — Опыт распределенного инференса/обучения гигантских моделей (MoE, multi-node clusters). — Понимание профилирования GPU, работы attention-кернелов и KV-cache. — Практика дообучения моделей (RLHF, DPO или аналоги). — Сильный бэкграунд на PyTorch / Transformers. — Английский или русский язык на профессиональном уровне. Плюсом будет: — Опыт разработки CUDA / Triton кернелов. — Бэкграунд в CV (опыт ускорения инференса гигантских GenAI моделей). — Опыт работы в AI-стартапах (Character.ai, Replika, OpenAI и др.) или BigTech (Яндекс, Сбер, T-Bank, Nebius, Meta, Mistral). — Профильное математическое / CS образование (МГУ, МФТИ, ВШЭ, Бауманка или топовые зарубежные вузы). Условия: — Полный Remote и максимум свободы без корпоративного gatekeeping. — Скорость: идеи становятся live-экспериментами за дни, а не quarters. — 28 дней отпуска + 7 wellness-дней в год без больничных листов. — $1,000 на обустройство хоум-офиса раз в 3 года. — Софинансирование обучения, конференций (50%) и медицинских расходов (до $1,000/год). — Доступ к опциональной A-team программе (опционы / % от выручки) после 1 года работы.

About the Role: You'll help build the very AI-powered platform you use every day, with real autonomy over the features you ship. Wrike generates an enormous amount of rich, interconnected data, including work item content, relationships between entities, complex business processes, user interactions, and direct user input. The mission of the DS/ML department is to transform this data into intelligent, AI-powered capabilities that help users gain deeper insights, automate routine work, enhance productivity, and create a more seamless work experience. As a Data Scientist in the Work Intelligence Unit, you will play a key role in product development through the application of state-of-the-art AI and machine learning technologies with a particular focus on leveraging Large Language Models and other emerging technologies. You will work on both user-facing AI features and foundational intelligence platform that power intelligent experiences across the product. Your Impact: — Creating new and improving existing AI solutions — Involvement in the whole process of data-driven development from analyzing product requirements to launching the feature or system to production (with the help of MLOps specialists) and improving it based on results of monitoring and feedback — Communicating with fellow data scientists, product owners, and developers — Applying best software engineering practices to build a reliable codebase — Performing code reviews — Working in agile environment and presenting the results of your work on regular sprint reviews Your Qualifications: — 4+ years of experience in AI,  Data Science, and Machine learning — Experience in LLMs, LLM-based APIs, prompting techniques, metric-driven development of prompt-based features, agentic systems, and other related topics — Experience in building complex multipart and multistep LLM-based and ML systems — Proficiency in Python (including pure Python, and not only jupyter notebooks) — Experience in using reproducible research techniques — Experience in developing AI solutions for production usage — Full professional proficiency in English Standout Qualities: — Experience in AI and ML solutions related to one of the areas: social networks, recommendation systems, ranking, entity matching, process mining, system personalization, risk estimation — Confident statistical and machine learning skills — Experience in creating core AI systems that are foundational to the product and are used in many parts of it — Experience in classical software development and architecture Interpersonal skills: — Excellent interpersonal, written, and verbal communication skills required with the ability to work within all levels of an organization. — You are proactively pushing yourself to develop your skills and knowledge — Decision making and argumentation skills — Open to feedback and ready to react on it — Ability to work autonomously on a project level Our Work Style: — Code: Python, Jupyter, mypy, uv, black, flake8 — Data: pandas, SQL, BigQuery — GenAI & LLMs: Langchain, Azure OpenAI, Gemini, Vertex AI, Braintrust — LLM Platform: Versioned Prompt Registry(jinja2), Declarative LLM Feature Configs (multi-step workflows, model routing) — ML: scikit-learn, lightgbm, pytorch — Cloud: GCP, Dagster, MLFlow — Coding Agents: Claude Code, Codex, Cursor Team Dynamics: — As part of Wrike’s AI center of excellence — you will help scale AI feature development across teams by collaborating closely with product, engineering, design, research, and other departments across the organization. — You will explore new AI-driven opportunities, validate hypotheses through experimentation, and drive innovations from research prototypes to production-ready capabilities that deliver tangible value to millions of users. Benefits & Perks: — 5 Weeks of paid vacation — Sick Leave Compensation — 5 Paid Uncertified Sick Days — 2 weeks fully paid w/ medical certificate, additional — 4 weeks paid at 80% salary rate — Parental Leave (fully paid): 18 Weeks Maternity / 4 Week Paternity — 2 Volunteer Days — Meal Vouchers (CZK 220 per working day) — Annual Prague Travel Card (Lítačka) — Hybrid Working Model — Benefit budget with flexible options, including a MultiSport card, Canadian Medical membership, contributions to a pension savings plan and additional choices available through Benefit Plus What’s Next? — Introduction Call with a Recruiter — Technical interview — Cultural interview

Location: San Francisco; Toronto; New York; Montreal; Seattle Workplace: Hybrid ABOUT BASETEN Baseten powers mission-critical inference for the world's most dynamic AI companies, like Cursor, Notion, OpenEvidence, Abridge, Clay, Gamma, and Writer. By uniting applied AI research, flexible infrastructure, and seamless developer tooling, we enable companies operating at the frontier of AI to bring cutting-edge models into production. We're growing quickly and recently raised our $1.5B Series F https://www.baseten.co/blog/announcing-our-series-f/, led by Altimeter Capital, Conviction Partners, and Spark Capital. Join us and help build the platform engineers turn to ship AI products. THE ROLE We're looking for inference performance engineers who want to make the world's most demanding AI workloads run faster and more efficiently. You'll work across the stack, from the inference engine and runtime through scheduling, serving, and routing. Along the way you'll apply techniques like prefill/decode disaggregation, speculative decoding, and KV-cache management. You'll reason from first principles about where time and memory go, find what's holding performance back, and close the gap. Your work directly impacts how fast our customers' models run and how efficiently we serve them. This role is ideal for someone who thrives in a fast-paced startup environment and is eager to make significant contributions to the exciting field of LLM inference.   EXAMPLE INITIATIVES You'll get to work on these types of projects as an Inference Performance engineer: - Agentic Kernels in Production https://www.baseten.co/blog/agentic-kernels-in-production/ - How we built the new fastest API for GLM-5.2 https://www.baseten.co/blog/how-we-built-the-new-fastest-api-for-glm-52/ - Live draft model training for speculative decoding https://www.baseten.co/blog/live-draft-model-training-for-speculative-decoding/#eliminating-traditional-bottlenecks - Making Kimi K3 Tokenization 18x faster https://www.baseten.co/blog/making-kimi-k3-tokenization-18x-faster-for-million-token-agentic-workloads/ - The Baseten Inference Stack https://www.baseten.co/resources/guide/the-baseten-inference-stack/ - Driving model performance optimization https://www.baseten.co/blog/driving-model-performance-optimization-2024-highlights/ RESPONSIBILITIES - Implement and productionize cutting-edge inference techniques, working deep in runtime internals. That includes quantization, speculative decoding, KV-cache reuse, chunked prefill, LoRA, guided generation for structured outputs, and custom scheduling and routing algorithms. - Profile and optimize inference end to end, from kernel launch overhead and memory layout up to request scheduling, prefill/decode disaggregation, and cache-aware routing. Run cross-layer investigations, such as tracing a tail-latency regression from request timing through routing and batching down to a kernel. - Turn performance into cost savings. Improve tokens per GPU-hour, raise utilization, and give customers and internal teams clear latency/throughput/cost tradeoffs. - Bring up and tune new model architectures on new hardware quickly, often in the same week they're released. - Build benchmarking frameworks that measure real-world performance across model architectures, batch sizes, sequence lengths, and hardware configurations. - Contribute upstream to open-source inference engines (vLLM, SGLang, TensorRT-LLM), and partner closely with model, infrastructure, and customer-facing teams to ship wins.   REQUIREMENTS - Bachelor's, Master's, or Ph.D. degree in Computer Science, Engineering, Mathematics, or related field. - Experience with one or more general-purpose programming languages, such as Python or C++. - Familiarity with LLM optimization techniques (e.g., quantization, speculative decoding, continuous batching). - Strong familiarity with ML libraries, especially PyTorch, TensorRT, or TensorRT-LLM. - Demonstrated interest and experience in LLMs. - Deep understanding of GPU architecture.   NICE TO HAVE - Proficiency in enhancing the performance of software systems, particularly in the context of large language models (LLMs) - Contributed to vLLM, SGLang, TensorRT-LLM, or another inference engine. - Worked on large-scale distributed serving: autoscaling, load balancing, multi-region or multi-cloud capacity. - Written or optimized GPU kernels (CUDA, Triton, CUTLASS, or similar) - Worked on quantization (FP8/FP4, AWQ, GPTQ) or speculative decoding in production. - Deep understanding of software engineering principles and a proven track record of developing and deploying AI/ML inference solutions. BENEFITS - Competitive compensation, including meaningful equity - (U.S. only) 100% coverage of medical, dental, and vision insurance for employee and dependents - Flexible PTO policy including company wide Winter Break (our offices are closed from Christmas Eve to New Year's Day!) - Paid parental leave - Fertility and family-building stipend through Carrot - (U.S. only) Company-facilitated 401(k) - Exposure to a variety of ML startups, offering unparalleled learning and networking opportunities. Apply now to embark on a rewarding journey in shaping the future of AI! If you are a motivated individual with a passion for machine learning and a desire to be part of a collaborative and forward-thinking team, we would love to hear from you. At Baseten, we are committed to fostering a diverse and inclusive workplace. We provide equal employment opportunities to all employees and applicants without regard to race, color, religion, gender, sexual orientation, gender identity or expression, national origin, age, genetic information, disability, or veteran status. We are an Equal Opportunity Employer and will consider qualified applicants with criminal histories in a manner consistent with applicable law (by example, the requirements of the San Francisco Fair Chance Ordinance, where applicable).

Location: San Francisco, CA • New York, NY Figma is growing our team of passionate creatives and builders on a mission to make design accessible to all. Figma’s platform helps teams bring ideas to life—whether you're brainstorming, creating a prototype, translating designs into code, or iterating with AI. From idea to product, Figma empowers teams to streamline workflows, move faster, and work together in real time from anywhere in the world. If you're excited to shape the future of design and collaboration, join us! The AI Research team at Figma is working on the cutting edge of AI and Design. We work on optimizing agentic systems for design quality and end user experience. We build robust evaluators for Design and post-train models. We are a very collaborative, close-knit and impact-focused team of researchers and research engineers working on some of the most challenging technical problems at the intersection of AI and creativity. We are seeking PhD interns with expertise in machine learning and artificial intelligence to develop innovative AI solutions for design and creative products. Through both fundamental and applied research, your work will contribute to new AI-driven features in Figma products. This internship can take place in Winter 2027 or Summer 2027 in San Francisco or New York. What you’ll do at Figma: Drive fundamental and applied research in Generative AI with Figma product use cases in mind Formulate and implement new modeling approaches both to improve the effectiveness of Figma’s current AI models as well as enable the launch of entirely new AI-powered product features Explore the boundaries of what is possible with the current technology set and experiment with novel ideas Depending on your project, you may have the opportunity to publish your research in leading AI/ML conferences We'd love to hear from you if you have: Are a 4th or 5th year senior Ph.D student in AI, Machine Learning, Computer Vision, Natural Language Processing, Human Computer Interaction or a related field Proven expertise in AI through publications on recent advancements in AI, such as Large Language Models or Agentic systems Track record in applying machine learning techniques to projects in academia or industry Experience in Python and machine learning frameworks (such as PyTorch, TensorFlow) Experience solving sophisticated problems and comparing alternative solutions, trade-offs, and diverse points of view to determine a path forward While it’s not required, it’s an added plus if you also have: Related industry experience via a past internship or full-time role Experience communicating your solutions effectively to different audiences Experience using AI-assisted development tools (e.g., GitHub Copilot, ChatGPT, Claude) to write, debug, and optimize code At Figma, one of our values is Grow as you go. We believe in hiring smart, curious people who are excited to learn and develop their skills. If you’re excited about this role but your past experience doesn’t align perfectly with the points outlined in the job description, we encourage you to apply anyways. You may be just the right candidate for this or other roles. Pay Transparency Disclosure This internship role is based in either Figma’s San Francisco or New York hub offices, and has the hourly base pay rate stated below.  Figma also offers interns a housing stipend and travel reimbursement. Figma’s compensation and benefits are subject to change and may be modified in the future. Internship $74 — $74 USD At Figma we celebrate and support our differences. We know employing a team rich in diverse thoughts, experiences, and opinions allows our employees, our product and our community to flourish. Figma is an equal opportunity workplace - we are dedicated to equal employment opportunities regardless of race, color, ancestry, religion, sex, national origin, sexual orientation, age, citizenship, marital status, disability, gender identity/expression, veteran status , or any other characteristic protected by law. We also consider qualified applicants regardless of criminal histories, consistent with legal requirements. We will work to ensure individuals with disabilities are provided reasonable accommodation to apply for a role, participate in the interview process, perform essential job functions, and receive other benefits and privileges of employment. If you require accommodation, please reach out to accommodations-ext@figma.com . These modifications enable an individual with a disability to have an equal opportunity not only to get a job, but successfully perform their job tasks to the same extent as people without disabilities. Examples of accommodations include but are not limited to: Holding interviews in an accessible location Enabling closed captioning on video conferencing Ensuring all written communication be compatible with screen readers Changing the mode or format of interviews To ensure the integrity of our hiring process and facilitate a more personal connection, we require all candidates keep their cameras on during video interviews. Additionally, if hired you will be required to attend in person onboarding. By applying for this job, the candidate acknowledges and agrees that any personal data contained in their application or supporting materials will be processed in accordance with Figma's Candidate Privacy Notice . Application limits: To help candidates focus on the opportunities most relevant to their skills and experience, Figma limits the number and frequency of applications. These limits apply to all roles: Candidates may submit up to 5 applications during any rolling 30-day period. Candidates may not reapply to the same role while an application is active or within 90 days of receiving a rejection. Once the applicable days has passed, candidates are welcome to apply again.

Location: Kyiv; Lviv Workplace: onsite Ajax Systems, an international tech company and the largest manufacturer of security systems in Europe.We are looking for a new team member to join our Video Devices Division as a Computer Vision Engineer . The Video Devices Division is responsible for the full development cycle of Ajax video devices, covering every stage from initial concept creation to mass production launch and further scaling. The team focuses on building complex products that combine mechanical design, hardware development, and the integration of low-level software to ensure stable real-time system performance.The development process involves deep expertise in optics, image quality, embedded systems, Wi-Fi and networking technologies, as well as audio tuning. The division ensures full vertical integration of all device components, creating technologically complete solutions that meet the reliability standards of the Ajax ecosystem and the requirements of global industrial production. What you'll work with: object detection; object tracking; face recognition; multi-modal search; acoustic intelligence; models optimization for edge inference; research and deliver sota computer vision solutions to exceed today's cctv domain expectations. Your cross-functional partners: ML team; platform team; annotation team; QA team. What we're looking for: 5+ years with machine learning; 2+ years with computer vision and python; experience with PyTorch ot TensorFlow. Nice to have: experience with C++; deploy AI models on edge devices; computer vision analytics troubleshooting. We offer: -the opportunity to suggest and implement your own ideas that directly influence the product and its mass production; - a place in a fast-growing company with a motivated and professional team; - competitive salary; - medical insurance; - corporate English classes; - the opportunity to participate in Ajax’s internal beta-testing program for security systems; - discounts from partner companies; - discounts on Ajax products. Ajax Systems is a Ukrainian success story that has grown into a global product. Ajax devices are trusted by more than 4.5 million end users and 330,000 PRO users in over 180 countries. The company provides comprehensive solutions to protect residential and commercial properties of all sizes. Today, Ajax portfolio includes 280 devices for intrusion protection, video surveillance, fire safety, and comfort and automation.

Задачи: — Исследовать большие объемы данных о покупательском поведении и находить закономерности для решения бизнес-задач — Разрабатывать и улучшать ML-модели для прогнозирования оттока и отклика пользователей — Решать задачи uplift-моделирования и выбора оптимального воздействия на пользователя — Подбирать персональные товары, категории, скидки, пороги и механики промо — Генерировать и проверять гипотезы, участвовать в подготовке и анализе A/B-тестов — Оценивать качество моделей не только по ML-метрикам, но и по влиянию на бизнес-показатели — Участвовать в развитии внутренних ML-инструментов и фреймворков команды — Участвовать в подготовке моделей к промышленному запуску и их дальнейшем развитии Требования: — Учишься на старших курсах бакалавриата, в магистратуре или недавно окончил техническую специальность — Хорошо знаешь Python и умеешь писать понятный и структурированный код — Знаешь основы машинного обучения и понимаешь принципы работы классических ML-алгоритмов — Умеешь работать с данными и знаешь SQL — Понимаешь основы математической статистики, теории вероятностей и проверки статистических гипотез — Понимаешь принципы проведения A/B-тестов и оценки результатов экспериментов — Умеешь самостоятельно разбираться в новой задаче, формулировать гипотезы и проверять их на данных — Интересуешься применением машинного обучения для решения реальных бизнес-задач Будет плюсом: — Знаком с PySpark и понимаешь основные принципы распределенной обработки данных — Работал с PyTorch или другими библиотеками для глубокого обучения — Знаком с uplift-моделированием, рекомендательными системами или задачами персонализации — Имеешь опыт участия в ML-проектах, соревнованиях, исследовательских или учебных проектах — Знаком с Airflow, Docker, Kubernetes, Grafana, MLflow — Имеешь представление о жизненном цикле ML-модели и процессе вывода моделей в промышленную эксплуатацию Условия: — Полная занятость — График: 5/2 — Формат работы: удалённо

Location: Berlin ROLE HIGHLIGHTS: - Senior MLOps Engineer, ISR - Location: Berlin, Germany - Department: ISR Analytics - Reports to: VP of ISR Analytics - Employment type: Permanent - Workplace model: Hybrid - Employment is subject to applicable security screening (incl. SUPO, where required) WHY THIS ROLE MATTERS: ICEYE's Intelligence, Surveillance and Reconnaissance (ISR) AI team builds models that turn satellite imagery into intelligence for defense customers. Your job is to work with data scientists to build sustainable infrastructure, workflows, and deployment practices for the AI models, both in the cloud and for on-premise systems. You sit with the AI team and work with them to build the infrastructure they train, evaluate and ship on. Like our platform work, every environment is temporary and must be reproducible from code; a model is only done when it runs, monitored, on hardware we don't control.   WHO WE ARE ICEYE is the world leader in sovereign intelligence from space. We deliver persistent monitoring capabilities to detect and respond to changes in any location on Earth.   ICEYE owns the world's largest and most advanced SAR (synthetic aperture radar) satellite constellation. To our customers we provide intelligence with unmatched quality, latency and revisit times, in any weather, day or night. To governments who choose to operate their own constellation we provide this proven capability as a sovereign system.   ICEYE-built constellations serve customers in defence and intelligence, environmental monitoring, insurance and emergency management. We enable fast decisions that contribute to a safer future.   Founded and headquartered in Finland, ICEYE operates globally with over 1000 employees across Europe, North America, the Middle East, and Asia-Pacific. YOUR DAY-TO-DAY RESPONSIBILITIES   - Build training, evaluation and packaging pipelines that take a model from experiment to versioned, deployable release. - Deploy and serve models on Kubernetes with GPUs, in the cloud and on-prem, including air-gapped sites. - Package models, weights and dependencies so they install and upgrade offline. - Manage GPU compute: scheduling, drivers, utilization and cost, across cloud and on-prem hardware. - Track data, experiments and model lineage, so any result can be reproduced and audited. - Monitor models in production for performance, drift and failures, without relying on outside connectivity. - Write tooling in Python that the AI team uses every day, and work in the model codebase alongside them.   WHAT WE’RE LOOKING FOR Must haves: - Senior hands-on engineer (not an engineering manager or architect role). - Proven experience deploying ML models to production and running them there, not only training them. - Strong Python, and comfort reading and changing ML code (PyTorch or similar). - Hands-on Kubernetes and containers, including GPU workloads. - Infrastructure as code and CI/CD for ML (for example Terraform, Helm, GitHub Actions). - Daily use of AI tools in engineering work, beyond chat: generating and reviewing code, configuration and tests. - High autonomy: you find problems, propose fixes and drive them through. - Motivation to work in new defense: building technology used by defense forces.   Nice to haves: - Deployed models to on-prem, edge or air-gapped environments. - Model serving and optimization: Triton, TorchServe, KServe, ONNX, TensorRT, quantization. - Computer vision or geospatial data (satellite, SAR, raster, PostGIS). - ML pipeline and tracking tools such as MLflow, Kubeflow, Argo Workflows, DVC or Weights & Biases. - Large-scale data processing for imagery (Dask, Ray, Spark, object storage). - Strong IC track record in a startup or a fast-changing company launching new products. - Clear communicator who works well with researchers and experienced engineers.   WORKING AT ICEYE At ICEYE, you’ll join a diverse and highly engaged team united by the ambition to make the impossible possible. As a global scale-up, we combine speed and ambition with the opportunity to take real ownership from day one. Your growth, wellbeing, and success are a priority, with continuous professional development, training opportunities, and a culture where collaboration is how we win. HOW WE WORK (OUR VALUES) Make the impossible possible: We set ambitious goals and stay calm under pressure. We bring grit, optimism, and ownership when things get hard, and we keep moving until we find a way. - Be curious: Go deep, ask questions, listen carefully, and think critically. Understand the “why” behind decisions. - See the big picture: Stay close to what’s happening across the company so you can make better decisions. Consider how your work affects others. - Drive effective teamwork: Create psychological safety, invite different perspectives, and build inclusive teams. There are no bad questions. - Act as one team: We win together. We match tasks to the right owner and stay agile as priorities shift. - Have fun: What we do matters—and it should be enjoyable. Celebrate progress, take pride in results, and share the wins. BENEFITS Our benefits are designed to support your health and wellbeing, at work and beyond. We keep improving them based on employee feedback, and offerings vary by location. Talent Acquisition will confirm what applies for this role and location during the process.   OUR COMMITMENT TO DIVERSITY, EQUITY, AND INCLUSION We want ICEYE to be a place where people can be themselves and do great work. Different backgrounds and perspectives make us stronger, which is why we work to create an environment where people feel included, respected, and able to speak up. Whatever your background, we want you to bring your authentic self to the table. We’re committed to fair, inclusive hiring and equal opportunity. Everyone is welcome to apply. If you need any adjustments or support during the recruitment process, tell us—we’ll do our best to help. Apply now to start your ICEYE journey, and help us continue to make the impossible possible together. Read more about ICEYE and working with us at iceye.com http://iceye.com.

Location: Москва Мы новое подразделение Сбера, занимающееся созданием инструментов и решений для внутренней безопасности с помощью AI. Наши задачи направлены на развитие AI-агентов, создание RAG, запуск локальных инстансов LLM и их файнтюнинг. Наши планы имеют широкий спектр - начиная от разработки DL-моделей для скоринга, поведенческого моделирования и заканчивая графовыми нейронками для выявления скрытых связей. Ищем Middle/Senior Data Scientist, который будет разрабатывать AI-агентов, адаптировать LLM под специфику задач, строить RAG-системы. Если ты хочешь иметь возможность погрузиться в глубины LLM и Deep Learning, пробовать новые методы к файнтюнингу LLM, строить AI-агентов, то мы ждем тебя в нашу команду. Обязанности * cоздание и развитие решений с задействованием LLM-агентов (полноценные мультиагентские системы) * cоздание RAG-систем * разработка и поддержка ML/DL моделей, включая их обновление и адаптацию к меняющимся условиям и требованиям бизнеса * дообучение и адаптация открытых LLM * создание системы мониторинга качества моделей и контроль их эффективности. Требования * оконченное высшее образование * опыт работы от пяти лет в коммерческой разработке на позиции Data Scientist / ML Engineer с подтверждённым опытом вывода моделей в продакшен * понимание математики классического ML и архитектуры нейросетей (PyTorch/TF) * уверенное понимание архитектуры трансформеров, опыт работы с LLM API и локальными моделями через PyTorch / Hugging Face * уверенный Python (чистый код, ООП, знание numpy/pandas/scikit-learn) и понимание принципов микросервесной архитектуры (Asyncio, FastAPI) * опыт создания агентов на LangChain/LangGraph * знание основных концепций промпт-инжиниринга и использования Structured Outputs и Function Calling * опыт проектирования и развития RAG-систем (Multi-stage Retrieval, Dense + Sparse/BM25, Rerankers, Chunking strategies) * опыт развертывания локальных open-source моделей (Llama, Mistral, Qwen, DeepSeek и др.) в продакшн-окружении и настройки высокопроизводительного инференса LLM с помощью vLLM * опыт проведения Fine-tuning LLM-моделей (PEFT, LoRA/QLoRA) * настройка мониторинга data drift и model decay, а также валидация моделей * навыки работы с генеративными AI-моделями; опыт создания AI-агентов и использования их в работе будет преимуществом * опыт использования GigaChat, Kandinsky и аналогов в продуктах, навыки создания и использования AI-агентов * инструментальное владение AI для анализа, генерации и автоматизации. Условия * инновационные, амбициозные проекты и задачи, которые развивают: всегда есть возможность прокачать свои навыки в работе и профессионально расти * среда для обмена знаниями – высокая экспертиза внутри команды; * сплоченная команда, работающая над общими задачами и умеющая хорошо отдыхать * нашу культуру создают сами сотрудники – мы их слышим и помогаем создавать и поддерживать корпоративные комьюнити по интересам * стабильная заработная плата и годовой бонус * полностью офисный формат работы. Современный IT-офис вблизи Москва-Сити в пяти минутах от метро "Кутузовская", с фитнес залом * более 400 образовательных программ СберУниверситета для профессионального и карьерного развития * подписка Прайм с возможностью совместного использования на трёх близких * расширенный ДМС, льготное страхование для семьи и корпоративная пенсионная программа * вознаграждение за рекомендацию друзей в команду Сбера.

Задачи: — Разработка и обучение моделей для анализа изображений и облаков точек — Подготовка датасетов, анализ данных — Сбор, анализ и предобработка данных, составление датасетов — Обучение моделей и тестирование гипотез — Деплой полученных решений — Написание технической документации — Презентация результатов внутри команды и заказчику — Изучение новых подходов для решения поставленных задач — Взаимодействие с командой разработки и аналитиками Требования: — Опыт разработки ML-решений для бизнеса от 3 лет — Опыт работы с CV на реальных проектах от 1 года — Опыт промышленной разработки на Python от 3 лет — Умение писать production-ready код на Python — Опыт работы с PyTorch — Умение обучать и дообучать модели, проводить тестирование — Опыт работы с Git — Умение версионировать данные и логировать эксперименты (DVC, MLFlow, ClearML) — Опыт работы с Docker — Опыт проработки требований, опыт декомпозиции и приоритизации задач — Умение вести техническую документацию — Знание английского B2 — Опыт работы с трехмерными данными Будет плюсом: — Опыт работы с ГИС и специальными форматами данных (geotiff, las)

Responsibilities: — Develop and adapt computer vision/video analytics models for detection, tracking, and recognition. — Work with video pipelines, including RTSP/HTTP, decoding, sampling, and batching. — Fine-tune models for real-world conditions and customer scenarios. — Optimize CPU inference, latency, and resource usage. — Integrate models into backend, edge, and on-premise solutions. — Prepare datasets, build training/testing pipelines, and monitor quality. Requirements: — 4+ years in Computer Vision/Video Analytics. — Strong CNN, YOLO, and SSD knowledge. — PyTorch or TensorFlow, OpenCV, and ONNX. — Understanding of RTSP, H.264/H.265, FPS, bitrate, and latency. — CPU inference optimization experience. — Linux and OpenVINO. — C++ for performance-critical components.

Задачи: — Разработка масштабной системы видеомониторинга и интеллектуальной видеоаналитики — Техническое лидерство направления Computer Vision — Ответственность за архитектуру AI-решений и развитие инженерной команды — Работа с production-системами, GPU-инференсом и Multimodal AI — Техническое руководство командой Computer Vision Engineers — Менторинг, code review и развитие инженерных практик — Проектирование архитектуры production-систем видеоаналитики — Разработка и оптимизация обработки RTSP-видеопотоков — Работа с NVIDIA DeepStream, GStreamer, TensorRT и Triton Inference Server — Внедрение моделей YOLO, ViT, DETR, Swin и Video Transformers — Оптимизация инференса через ONNX, TensorRT, OpenVINO, CUDA и Quantization — Разработка AI-сервисов на FastAPI / gRPC — Работа с Docker и Kubernetes — Участие в найме и технической оценке специалистов Требования: — Коммерческий опыт в Computer Vision / Video Analytics — Опыт технического лидерства или наставничества — Опыт запуска production AI-систем — Знание Python, PyTorch, OpenCV, YOLO — Опыт работы с RTSP, GStreamer, NVIDIA DeepStream, TensorRT — Опыт с ONNX, OpenVINO, CUDA или Quantization — Понимание высоконагруженных систем обработки видеопотоков — Опыт с Docker и Kubernetes — Знание Vision Transformers и мультимодальных моделей Будет преимуществом: — Опыт работы с Triton Inference Server — Знание C++ / CUDA — Опыт работы с Kafka — Знание MLflow / MLOps — Опыт с VLM, RAG, LangGraph, vLLM, Function Calling — Опыт формирования AI-команд Условия: — Полная занятость — График 5/2 — Офисный формат — Заработная плата 700 000–1 700 000 ₸ до вычета налогов — Возможность влиять на архитектуру и технологический стек — Работа над масштабной production-системой Адрес: Астана, улица Сыганак, 60/2

Задачи: — Владеть и внедрять AI-функции в продакшн, используя LLM и собственные модели. — Обучать специализированные модели для задач, таких как модерация контента и рекомендации. — Сотрудничать с продакт-менеджерами для превращения целей в масштабируемые решения. — Оценивать и принимать решения по стратегиям разработки или покупки на основе качества и стоимости. — Мониторить производительность развернутых моделей и вносить необходимые корректировки. Требования: — Опыт работы инженером 5+ лет, с практическим опытом работы с LLM и обучением моделей. — Подтвержденная способность развертывать модели под реальной нагрузкой и оптимизировать задержки. — Сильные навыки Python и PyTorch, а также твердые знания в области машинного обучения. — Опыт мониторинга и версионирования моделей в продакшне. — Владение английским языком на уровне Intermediate или выше. Плюсом будет: — Опыт работы с инференсом в реальном времени на живом видео/аудио. — Знакомство с рекомендательными системами и мультимодальной работой. — Знание Docker, Kubernetes и облачных сред. Условия: — Конкурентная зарплата, бонусы и опционы (все являются партнерами). — Официальное трудоустройство на Кипре с поддержкой визы для сотрудника и супруга. — Офис в центре Лимассола с питанием и корпоративной культурой. — Медицинская страховка (100% для сотрудника, 75% для семьи). — Доступ к матчам и мероприятиям Aris Limassol F.C. — Возможности профессионального роста и корпоративные семейные мероприятия. Формат работы: в офисе (Лимассол).

Location: San Francisco, CA 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 Niantic Spatial makes the physical world computable, helping people and machines collaborate safely by aligning how they understand reality. One of the fundamental problems we're helping autonomy teams and engineers overcome is the sim-to-real gap for visual-spatial understanding. We're applying our team's decades of experience encoding the world precisely as it is at Google Maps, Google Earth, and Niantic Labs to build the scalable real-to-sim stack for embodied AI. We're looking for a Staff Engineer, Robot Learning to lead engineering for our real-to-sim product within the Embodied AI group. One of the fundamental problems facing autonomy teams is the sim-to-real gap in visual-spatial understanding, and we're applying decades of experience encoding the world precisely as it is to build the scalable real-to-sim stack for embodied AI. Working alongside the group's General Manager and Product leader, you'll inform the roadmap, own the technical requirements our researchers and engineers build against, and personally prototype and evaluate the changes you propose. You've lived the problems our customers are solving: you've trained and deployed embodied AI systems, seen where and why simulation fails to predict real-world performance, and worked through the consequences. This is not a role for someone who wants to manage and not ship, or someone who wants to ship without ownership of commercial consequences. This is a role for an entrepreneurial tech lead who loves to build and is energized by helping those around them do the best work of their lives. You have a co-founder mentality. You may have been part of a founding team before, or want to start your own business one day. What You'll Do - Sharpen the Roadmap - Bring first-hand knowledge of customer pain, existing solutions, and the value of doing better to roadmap decisions, and help focus effort on the most consequential problems to solve. - Own the Technical Requirements - Guide the team to design the real-to-sim stack pragmatically. Make the final recommendation on what to build, integrate, and simplify, ensuring capabilities work within customers' existing training and evaluation workflows. - Benchmark the Impact - Working with our Real-World Test Lab, design and run reproducible evaluations that measure the impact of our real-to-sim stack on policy transfer and real-world performance against customers' existing approaches. - Build and Ship - Lead engineering delivery and personally build critical features. Keep scope tight, remove blockers, and get performant capabilities into customers' hands quickly. - Raise the Technical Bar - Set a high bar for the team's work and deepen its understanding of robotics development and deployment, driving incremental improvement every week. What You'll Bring - Experience training, evaluating, and deploying robot policies in systems used by enterprise or government customers, with responsibility for performance in real operating conditions. - Deep practical experience using simulation for policy training or evaluation, including tools such as NVIDIA Isaac Sim, Isaac Lab, MuJoCo, or equivalent. You have diagnosed sim-to-real failures and understand when greater fidelity improves outcomes. - Hands-on experience with reinforcement learning, imitation learning, or both, and with deep-learning frameworks such as PyTorch. - A practical understanding of how visual perception and 3D scene representations affect robotics training, evaluation, and real-world performance. - Experience running experiments on real robots, from an ambiguous hypothesis through to defensible results. - Strong Python and C++ skills, with experience building reliable production software. - A bachelor's degree or higher in a relevant field. Nice to Have - You've authored or maintained a public robotics benchmark. - You've helped take an early product to repeatable adoption across multiple customers. - You've built or substantially extended robotics simulation infrastructure, including scene authoring, domain randomization, or large-scale training and evaluation. - You've trained and deployed vision-based navigation or mobility policies, particularly in varied, unstructured environments. - You've built pipelines that turn real-world reconstructions into simulation-ready environments. Competencies - Always Learning. Staying on top of the latest academic breakthroughs and commercial advancements, particularly in real-to-sim-to-real. - Relentless bias for action. You make decisions and ship as if the company’s success depends on it, because it does. You set the pace, and drive the people around you to rise to it. - Intellectually honest. You know there’s just one job that underpins every other in a startup: find the truth. You are relentless in your pursuit of it, including when it challenges your own convictions, and you raise concerns when you see them — even when it's uncomfortable, and especially when it's unpopular. - Pragmatic. You have no patience for not-invented-here syndrome and analysis paralysis. You find the fastest path to delivering customer value without going through one-way doors that undermine future scaling. - Confident and credible with customers. You can sit alongside an autonomy engineer and help them get to the root of their problems. You are comfortable speaking to customers and bring commercial nous to those discussions. Compensation & Benefits The expected salary range for this role is $270,000 – $300,000 per year. Compensation also includes an annual bonus, equity, and a comprehensive benefits package including medical, dental, and vision coverage, 401(k), and more. Location & Work Model This role is based in our San Francisco 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: Berlin adjoe builds the technologies behind mobile app growth and monetization, challenging the status quo of adtech by redefining how apps acquire users and generate revenue. Our two products, Playtime Arcade and adjoe Ads, sit at opposite ends of the mobile advertising spectrum: rewarded engagement and programmatic in-app advertising. Together they run on 13% of the world’s smartphones, connecting advertisers, publishers, and hundreds of millions of users, with one goal: making mobile advertising work better for everyone in the ecosystem. Next-Gen Machine Intelligence for Auctions & Real-Time Bidding The Data Science Research team builds the foundational algorithms that govern real-time bidding, budget pacing, and allocation mechanics across our products. Operating as a horizontal deep-tech layer, the team leverages advanced AI—specifically reinforcement learning, objective-based optimization, and control theory—to deliver critical systems that perfectly harmonize advertiser ROI, platform health, and user relevance. These frameworks scale adjoe’s products beyond standard industry limitations, utilizing behavioral and transactional data from over 770 million users. The team operates at the intersection of high-load, low-latency engineering, auction mechanism design, and state-of-the-art machine learning. Acting as the company’s internal science bar raiser, they translate complex business constraints into mathematical objectives. The stack is optimized for high throughput: PyTorch for training deep learning and RL models, distributed architectures for large-scale Bayesian inference, Trino and Athena for querying massive user-behavior data lakes, and dedicated GPU infrastructure. The full lifecycle of these AI capabilities is owned by the team—from academic paper translation and offline prototyping to deploying core algorithmic frameworks built to sustain ultra-low latency, real-time production environments evaluating tens of thousands of requests per second. Your Mission & Who We Are Looking For: - PhD or equivalent research depth, with a production track record. You have a PhD in CS, Statistics, Math, or Physics (or an MSc with extensive industry research) and a track record of peer-reviewed publications in fields like Computational Advertising, Auction Theory, Mechanism Design, or Game Theory. You’ve spent 4+ years optimizing complex systems through automated algorithms in AdTech, marketplaces, or high-frequency trading. - You translate research into production, not just prototypes. You distill SOTA research into robust, production-ready code optimized for sub-millisecond latency. At adjoe, that means architecting sophisticated auction mechanics, real-time bidding engines, and allocation algorithms serving millions of requests across a stack built on 770 million users. - Deep expertise in Reinforcement Learning and Optimization. You have hands-on, deep theoretical and technical expertise in objective-based optimization, control theory, or RL applied specifically to bidding, pricing strategy, and budget pacing under extreme production constraints. - Mastering high-dimensional data at scale. Large-scale Bayesian inference and handling sparse, high-dimensional advertising data are core to your work. You are strongly proficient in Python and related stacks, and know exactly how to architect scalable data pipelines to feed real-time optimization models. - Raises the technical bar around them. You lead architectural debates, collaborate tightly with Product Managers and Engineers to map business goals to mathematical objective functions, mentor junior scientists, and drive the team’s research direction without needing a roadmap handed to you. 🌎 WE WELCOME APPLICATIONS FROM TALENT WORLDWIDE AND PROVIDE RELOCATION SUPPORT TO HAMBURG, GERMANY FOR THOSE READY TO JOIN OUR TEAM. What’s in It for You? At adjoe, you’re not here to just close JIRA tickets, you’re helping build the infrastructure behind one of the most impactful platforms in adtech. The systems you work on will reach hundreds of millions of users and power billions of decisions every day. - Go Big. Own projects with impact on 770M users and push adtech boundaries. - Move Fast. Ship solutions multiple times a day, learn from results, and keep momentum. - Be Direct. Solve problems openly and collaborate across teams. - Thrive Together. Grow with a diverse, global team of people from over 40 different countries that learn from each other. - Have Fun. Celebrate wins, enjoy daily victories, and bring your energy. We welcome applications from people who will contribute to the diversity of our company.

Location: Москва Мы — core-команда, которая отвечает за ML для понимания аудио во всем Сбере. За последний год мы выложили в open source SOTA-модель для распознавания речи на русском языке — **GigaAM**, а прошлой весной первыми в России запустили **нативное понимание звука в LLM — GigaChat Audio**. Сейчас мы активно улучшаем мультимодальные возможности GigaChat: работаем над качеством на сложных аудио- и визуальных контекстах и учим модель понимать видео не только по звуковому потоку, но и по кадрам. **Цель** — совместное обучение на vision+audio и video+audio диалогах длительностью 90+ минут. _Первый этап отбора на эту вакансию – общение с AI-рекрутером. После отклика вам на почту и в чат на платформе HeadHunter придет приглашение пройти первичное интервью с ГигаРекрутером в Telegram. Диалог займёт примерно 10 минут. Его задача — уточнить недостающие детали и ускорить рассмотрение вашей кандидатуры. ГигаРекрутер только начинает свой путь, поэтому просим относиться с пониманием. Ваш опыт и участие помогут сделать его удобным и полезным!_ Обязанности * Проводить эксперименты по обучению Vision / Audio / Video LLM: данные, стадии обучения, способы смешивания модальностей * Оптимизировать ML-инфраструктуру для обучения на длинных мультимодальных контекстах * Формулировать и проверять гипотезы о том, как улучшать качество LLM на видео. Требования * Сильный опыт в DL-инженерии, желательно в обучении LLM-моделей * Глубокое знание PyTorch, включая DDP / FSDP * Понимание distributed training и efficient deep learning: DP / TP / PP / EP / SP, mixed precision, checkpointing, offloading, профилирование и оптимизация обучения * Понимание архитектуры LLM: Transformer, attention (MHA / GQA / MLA), RoPE и другие позиционные эмбеддинги, long context, MoE * Уверенный Python на уровне production-кода: asyncio, multiprocessing, профилирование, отладка больших систем. **Будет плюсом:** * Опыт обучения мультимодальных LLM * Опыт снижения стоимости контекста при обработке видео * Опыт расширения контекста LLM * Понимание RL-подходов для обучения моделей: RLHF / RLVR, PPO / GRPO / DPO * Опыт построения бенчмарков с использованием LLM-as-a-judge. Условия * крупнейшее DS&AI community — более 600 DS-специалистов банка * дайджест о самых последних разработках в области DS&AI и отчеты с крупнейших конференций мира * возможность быть соавтором НИРов и статей для международных конференций * возможность выбрать удобный формат работы: гибрид или офис * ежегодный пересмотр зарплаты, 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 * ипотека выгоднее до 7% для каждого сотрудника * бесплатная подписка СберПрайм+, скидки на продукты компаний-партнеров * вознаграждение за рекомендацию друзей в команду Сбера.