Research Engineer, Infrastructure

At Thinking Machines Lab, we are dedicated to empowering humanity by advancing collaborative general intelligence. Our vision is to create a future where everyone can leverage AI to meet their unique needs and aspirations.Our talented team comprises scientists, engineers, and innovators who have developed some of the most widely recognized AI products, including ChatGPT and Character.ai, alongside open-weight models like Mistral and popular open-source projects such as PyTorch, OpenAI Gym, Fairseq, and Segment Anything.About the PositionWe are seeking a motivated Infrastructure Research Engineer to design, enhance, and scale the systems that underpin large AI models. Your contributions will significantly improve inference speed, cost-effectiveness, reliability, and reproducibility, allowing our teams to concentrate on enhancing model capabilities rather than dealing with bottlenecks.Our mission centers on delivering high-performance and efficient model inference to support real-world applications and accelerate research efforts. In this role, you will be responsible for the infrastructure that guarantees smooth operation for every experiment, evaluation, and deployment at scale.Note: This is an evergreen role, kept open continuously to express interest. We receive numerous applications and may not always have an immediate opening that aligns perfectly with your skills and experience. However, we encourage you to apply. We regularly review applications and reach out to candidates as new opportunities arise. Feel free to reapply as you gain more experience, but we kindly ask that you avoid applying more than once every six months. You may also notice postings for specific roles related to particular projects or teams, in which case you are welcome to apply directly in addition to this evergreen role.What You Will DoCollaborate with researchers and engineers to transition cutting-edge AI models into production.Partner with research teams to ensure high-performance inference for innovative architectures.Design and implement new techniques, tools, and architectures that enhance performance, latency, throughput, and efficiency.Optimize our codebase and computing resources (e.g., GPUs) to maximize hardware FLOPs, bandwidth, and memory usage.Extend orchestration frameworks (e.g., Kubernetes, Ray, SLURM) for distributed inference, evaluation, and large-batch serving.Establish standards for reliability, observability, and reproducibility throughout the inference stack.Publish and share insights through internal documentation, open-source libraries, or technical reports that further the field of scalable AI infrastructure.

About Our TeamAt OpenAI, our Foundations team is dedicated to examining how model behavior evolves as we scale up models, data, and computing resources. We meticulously analyze the relationships between model architecture, optimization strategies, and training datasets to inform the design and training of next-generation models.About the PositionAs a Team Lead in Research Inference, you will be instrumental in constructing systems that empower advanced AI models to operate efficiently at scale. Your role lies at the crossroads of model research and systems engineering, where you will translate innovative architectural concepts into high-performance inference systems, clearly illustrating the trade-offs in performance, memory usage, and scalability.Your contributions will significantly shape model design, evaluation, and iteration processes across our research organization. By developing and refining high-performance inference infrastructures, you will provide researchers with the tools necessary to explore new ideas while understanding their computational and systems implications.This position does not involve serving products; instead, it supports research through a focus on performance, accuracy, and realism, ensuring that our AI research is firmly rooted in scalable solutions.ResponsibilitiesDesign and develop optimized inference runtimes for large-scale AI models, emphasizing efficiency, reliability, and scalability.Take ownership of optimizing core execution processes, including model execution, memory management, batching, and scheduling.Enhance and expand distributed inference across multiple GPUs, focusing on parallelism, communication patterns, and runtime coordination.Implement and refine critical inference operators and kernels based on real-world workloads.Collaborate closely with research teams to ensure accurate and efficient support for new model architectures within inference systems.Identify and resolve performance bottlenecks through comprehensive profiling, benchmarking, and low-level debugging.Contribute to the observability, correctness, and reliability of large-scale AI systems.Ideal Candidate ProfileExperience in developing production-level inference systems, beyond just training and executing models.Proficient in GPU-centric performance engineering, including managing memory behavior and understanding latency/throughput trade-offs.Strong analytical skills and familiarity with performance profiling tools.

Who are we?At Cohere, our mission is to elevate intelligence to benefit humanity. We specialize in training and deploying cutting-edge models for developers and enterprises focused on creating AI systems that deliver extraordinary experiences such as content generation, semantic search, retrieval-augmented generation, and intelligent agents. We view our work as pivotal to the broad acceptance of AI technologies.We are passionate about our creations. Every team member plays a vital role in enhancing our models' capabilities and the value they provide to our customers. We thrive on hard work and speed, always prioritizing our clients' needs.Cohere is a diverse team of researchers, engineers, designers, and more, all dedicated to their craft. Each individual is a leading expert in their field, and we recognize that a variety of perspectives is essential to developing exceptional products.Join us in our mission and help shape the future of AI!Why this role?Are you excited about architecting high-performance, scalable, and reliable machine learning systems? Do you aspire to shape and construct the next generation of AI platforms that enhance advanced NLP applications? We are seeking talented Members of Technical Staff to join our Model Serving team at Cohere. This team is responsible for the development, deployment, and operation of our AI platform, which delivers Cohere's large language models via user-friendly API endpoints. In this role, you will collaborate with multiple teams to deploy optimized NLP models in production settings characterized by low latency, high throughput, and robust availability. Additionally, you will have the opportunity to work directly with customers to create tailored deployments that fulfill their unique requirements.

OpenAI's research infrastructure group creates and maintains the backbone systems for advanced machine learning model training. This team often goes beyond conventional training methods, developing new infrastructure to support novel research at scale. Their work closely connects systems engineering with research progress, making it possible to run experiments that would otherwise be too slow or complex. Role overview The Research Infrastructure Engineer for Training Systems designs and improves the platforms that power large-scale ML training. This role bridges research concepts and the practical systems that make large model training possible. The work has a direct impact on model release timelines and requires building systems that perform reliably in demanding, real-world scenarios. What you will do Build and maintain infrastructure for large-scale model training and experimentation Design APIs and interfaces to simplify complex training workflows and prevent misuse Enhance reliability, debuggability, and performance across training and data pipelines Troubleshoot issues involving Python, PyTorch, distributed systems, GPUs, networking, and storage Create tests, benchmarks, and diagnostic tools to catch regressions early Requirements Interest in building systems that support new training methods, not just optimizing existing ones Strong instincts in systems engineering, especially regarding performance, reliability, and clean abstractions Experience designing APIs and interfaces for researchers and engineers Ability to work across ML research code and production infrastructure Enjoys evidence-based debugging using profiles, traces, logs, tests, and reproducible cases

At Inferact, we are on a mission to establish vLLM as the premier AI inference engine, revolutionizing AI progress by making inference both more accessible and efficient. Our founding team consists of the original creators and key maintainers of vLLM, positioning us uniquely at the nexus of cutting-edge models and advanced hardware.Role OverviewWe are seeking a passionate inference runtime engineer eager to explore and expand the frontiers of LLM and diffusion model serving. As models evolve and grow in complexity with new architectures like mixture-of-experts and multimodal designs, the demand for innovative solutions in our inference engine intensifies. This role places you at the heart of vLLM, where you will enhance model execution across a variety of hardware platforms and architectures. Your contributions will have a direct influence on the future of AI inference.

At Thinking Machines Lab, we are on a mission to empower humanity by advancing collaborative general intelligence. Our vision is to create a future where everyone has access to the knowledge and tools necessary to harness AI for their unique needs and objectives.We are a diverse team of scientists, engineers, and builders responsible for developing some of the most influential AI products on the market, such as ChatGPT and Character.ai. Our contributions extend to open-weight models like Mistral and popular open-source projects including PyTorch, OpenAI Gym, Fairseq, and Segment Anything.About the RoleWe are seeking talented engineers to join our team and develop the libraries and tools that will accelerate research efforts at Thinking Machines. You will take charge of our internal infrastructure—creating evaluation libraries, reinforcement learning training libraries, and experiment tracking platforms—while building systems that enhance research velocity over time.This position emphasizes collaboration. You will work closely with researchers to identify bottlenecks and pain points, ensuring that they trust your systems to function seamlessly and find them enjoyable to use.What You'll DoDesign, build, and manage research infrastructure, including evaluation frameworks, RL training systems, experiment tracking platforms, visualization tools, and shared utilities.Develop high-throughput, scalable pipelines for distributed evaluation, reward modeling, and multimodal assessment.Establish systems for reproducibility, traceability, and robust quality control across research experiments and model training runs, implementing effective monitoring and observability.Collaborate directly with researchers to identify bottlenecks and unlock new capabilities, managing research tools like a product manager by proactively seeking feedback and tracking adoption.Work alongside infrastructure, data, and product teams to integrate tools across the technical stack.

Join the Fluidstack TeamAt Fluidstack, we’re pioneering the infrastructure for advanced intelligence. We collaborate with leading AI laboratories, governmental entities, and major corporations—including Mistral, Poolside, and Meta—to deliver computing solutions at unprecedented speeds.Our mission is to transform the vision of Artificial General Intelligence (AGI) into a reality. Driven by our purpose, our dedicated team is committed to building state-of-the-art infrastructure that prioritizes our customers' success. If you share our passion for excellence and are eager to contribute to the future of intelligence, we invite you to be part of our journey.Role OverviewThe Inference Platform team at Fluidstack is at the forefront of addressing the cost and latency challenges associated with frontier AI. You will play a crucial role in managing the serving layer that connects our global accelerator supply with the production workloads of our clients, which include LLM serving frameworks, KV cache infrastructure, and Kubernetes orchestration across multiple data centers.This hands-on individual contributor role combines elements of distributed systems, model optimization, and serving infrastructure. You will oversee the entire lifecycle of inference deployments for leading AI labs, striving for enhancements in throughput, cost-efficiency, and response times, while also influencing the architectural decisions that guide Fluidstack’s deployment strategies.

At Thinking Machines Lab, we are committed to empowering humanity by advancing collaborative general intelligence. Our vision is to create a future where everyone has access to the knowledge and tools necessary to harness AI for their unique needs and aspirations.Our team comprises scientists, engineers, and builders who have developed some of the most utilized AI products, including ChatGPT and Character.ai, as well as open-weight models like Mistral. We also contribute to notable open-source projects such as PyTorch, OpenAI Gym, Fairseq, and Segment Anything.About the RoleWe are seeking a talented Infrastructure Research Engineer to enhance, scale, and fortify the systems supporting Tinker. This role will enable our internal teams and external clients to fine-tune models seamlessly, reliably, and cost-effectively. You will work at the intersection of large-scale training systems and product infrastructure, creating multi-tenant scheduling, storage, observability, and reliability features within a developer-friendly API.Your contributions will allow all Tinker users to concentrate on research and development without the burden of infrastructure concerns.Note: This is an evergreen position that we keep open for ongoing interest. We receive numerous applications, and there may not always be a role that aligns perfectly with your skills and experience. We encourage you to apply, as we continuously review applications and will reach out as new opportunities arise. You are welcome to reapply after gaining more experience, but please refrain from applying more than once every 6 months. We also post specific roles for unique project or team needs, and you are welcome to apply directly to those in addition to this evergreen listing.What You’ll DoDesign and implement distributed job orchestration, placement, preemption, and fair-share scheduling to enhance Tinker for multi-tenant workloads.Optimize GPU utilization, throughput, and reliability across clusters (including autoscaling, bin-packing, and quotas).Develop reusable frameworks and libraries to enhance Tinker’s transparency, reproducibility, and performance.Collaborate with researchers and developer experience engineers to transform fine-tuning challenges into product features.Publish and disseminate insights through internal documentation, open-source libraries, or technical reports to advance the field of scalable AI infrastructure.

At Thinking Machines Lab, our mission is to empower humanity by advancing collaborative general intelligence. We envision a future where everyone has access to the knowledge and tools necessary to harness AI for their unique needs and goals.Our team comprises scientists, engineers, and builders who have developed some of the most widely utilized AI products, such as ChatGPT and Character.ai, alongside open-weight models like Mistral, and popular open-source initiatives like PyTorch, OpenAI Gym, Fairseq, and Segment Anything.About the PositionWe are seeking an Infrastructure Research Engineer to design and construct the foundational systems that facilitate the scalable and efficient training of large models for both deployment and research purposes. Your primary objective will be to streamline experimentation and training at Thinking Machines, enabling our research teams to concentrate on scientific advancements rather than system limitations.This role is a perfect match for an individual who possesses a strong blend of deep systems expertise and a keen interest in machine learning at scale. You will take full ownership of the training stack, ensuring that every GPU cycle contributes to scientific progress.Note: This is an evergreen role that we keep open continuously to express interest. We receive numerous applications, and there may not always be an immediate role that aligns perfectly with your experience and skills. However, we encourage you to apply. We regularly review applications and reach out to candidates as new opportunities arise. Feel free to reapply as you gain more experience, but please avoid applying more than once every six months. We may also post specific roles for individual projects or team needs, in which case you are welcome to apply directly alongside this evergreen role.Key ResponsibilitiesDesign, implement, and optimize distributed training systems that scale across thousands of GPUs and nodes for extensive training workloads.Develop high-performance optimizations to maximize throughput and efficiency.Create reusable frameworks and libraries that enhance training reproducibility, reliability, and scalability for new model architectures.Establish standards for reliability, maintainability, and security, ensuring systems remain robust under rapid iterations.Collaborate with researchers and engineers to construct scalable infrastructure.Publish and disseminate findings through internal documentation, open-source libraries, or technical reports that contribute to the advancement of scalable AI infrastructure.

OverviewAt Pulse, we are revolutionizing the way data infrastructure operates by addressing the critical challenge of accurately extracting structured information from intricate documents on a large scale. Our innovative document understanding technique merges intelligent schema mapping with advanced extraction models, outperforming traditional OCR and parsing methods.Located in the heart of San Francisco, we are a dynamic team of engineers dedicated to empowering Fortune 100 enterprises, YC startups, public investment firms, and growth-stage companies. Backed by top-tier investors, we are rapidly expanding our footprint in the industry.What sets our technology apart is our sophisticated multi-stage architecture, which includes:Specialized models for layout understanding and component detectionLow-latency OCR models designed for precise extractionAdvanced algorithms for reading-order in complex document structuresProprietary methods for table structure recognition and parsingFine-tuned vision-language models for interpreting charts, tables, and figuresIf you possess a strong passion for the convergence of computer vision, natural language processing, and data infrastructure, your contributions at Pulse will significantly impact our clients and help shape the future of document intelligence.

About Our Innovative TeamJoin the Workload team at OpenAI, where we are at the forefront of designing and managing the cutting-edge infrastructure that drives the training and inference of large language models (LLMs) at an unprecedented scale. Our systems are engineered to harmonize the complex processes of model training and serving, abstracting performance, parallelism, and execution across extensive GPU and accelerator networks. This robust foundation allows researchers to concentrate on elevating model capabilities, while we take care of the scalability, efficiency, and reliability needed to bring these advanced models to life.Your Role and ResponsibilitiesWe are seeking a talented engineer to design and implement the dataset infrastructure that will fuel OpenAI’s next-generation training stack. Your primary focus will be on creating standardized dataset interfaces, scaling pipelines across thousands of GPUs, and proactively identifying and addressing performance bottlenecks. Collaboration with multimodal researchers and infrastructure teams will be key to ensuring that our datasets are unified, efficient, and user-friendly.Key Responsibilities Include:Design and maintain standardized dataset APIs, including those for multimodal (MM) data that exceeds memory capacity.Develop proactive testing and validation pipelines for dataset loading at GPU scale.Work collaboratively to integrate datasets into training and inference pipelines, ensuring seamless user experiences.Document and maintain dataset interfaces to ensure they are discoverable, consistent, and easily adoptable by other teams.Establish validation systems to assure datasets remain reproducible and unchanged once standardized.Identify and troubleshoot performance bottlenecks in distributed dataset loading, such as stragglers impacting global training speed.Create visualization and inspection tools to highlight errors, bugs, or bottlenecks in datasets.Ideal Candidate ProfilePossess strong engineering fundamentals and experience in distributed systems, data pipelines, or infrastructure.Have a proven track record in building APIs, modular code, and scalable abstractions, with a user-centric approach to design.Be adept at debugging performance issues across large-scale machine fleets.Demonstrate a passion for advancing data infrastructure to enhance research capabilities.

At Thinking Machines Lab, our mission is to empower humanity by advancing collaborative general intelligence. We envision a future where everyone has access to the knowledge and tools necessary to make AI work for their individual needs and goals. Our team comprises scientists, engineers, and innovators who have developed some of the most widely adopted AI products, including ChatGPT and Character.ai, alongside open-weight models like Mistral, as well as popular open-source initiatives such as PyTorch, OpenAI Gym, Fairseq, and Segment Anything.About the RoleWe are seeking a highly skilled infrastructure research engineer to architect and develop core systems that facilitate efficient large-scale model training, with a strong emphasis on numerics. You will enhance the numerical foundations of our distributed training stack, focusing on precision formats, kernel optimizations, and communication frameworks to ensure that training trillion-parameter models is stable, scalable, and fast.This position is perfect for an individual who excels at the intersection of research and systems engineering—a creator who comprehends both the mathematics of optimization and the practicalities of distributed computing.Note: This is an "evergreen role" that remains open for ongoing expressions of interest. While we receive numerous applications and there may not always be an immediate opening that perfectly matches your skills and experience, we encourage you to apply. We continuously review applications and will contact applicants as new opportunities arise. You are welcome to reapply if you gain additional experience, but please refrain from applying more than once every six months. You may also notice postings for specific roles related to particular projects or teams; in those instances, you are welcome to apply for those positions in addition to the evergreen role.What You’ll DoDesign and optimize distributed training infrastructure for large-scale LLMs, ensuring performance, stability, and reproducibility in multi-GPU and multi-node environments.Implement and assess low-precision numerics (e.g., BF16, MXFP8, NVFP4) to enhance efficiency while maintaining model quality.Develop kernels and communication primitives that leverage hardware-level support for mixed and low-precision arithmetic.Collaborate with research teams to co-design model architectures and training methodologies that align with new numeric formats and stability requirements.Prototype and benchmark scaling strategies, including data, tensor, and pipeline parallelism that integrate precision-adaptive computation and quantized communication.Contribute to the design of our internal orchestration and monitoring frameworks.

Join Cartesia as an Inference EngineerAt Cartesia, our vision is to create the next evolution of AI: an interactive, omnipresent intelligence that operates seamlessly across all environments. Currently, even the most advanced models struggle to continuously analyze a year's worth of audio, video, and text data—comprising 1 billion text tokens, 10 billion audio tokens, and 1 trillion video tokens—much less perform these tasks on-device.We are at the forefront of developing the model architectures that will make this a reality. Our founding team, who met as PhD candidates at the Stanford AI Lab, pioneered State Space Models (SSMs), a groundbreaking framework for training efficient, large-scale foundation models. Our talented team merges deep expertise in model innovation and systems engineering with a design-focused product engineering approach, enabling us to build and launch state-of-the-art models and user experiences.Supported by leading investors such as Index Ventures and Lightspeed Venture Partners, along with contributions from Factory, Conviction, A Star, General Catalyst, SV Angel, Databricks, and others, we are fortunate to be guided by numerous exceptional advisors and over 90 angel investors from diverse industries, including some of the world’s foremost experts in AI.About the RoleWe are actively seeking an Inference Engineer to propel our mission of creating real-time multimodal intelligence.Your ImpactDevelop and implement a low-latency, scalable, and dependable model inference and serving stack for our innovative foundation models utilizing Transformers, SSMs, and hybrid models.Collaborate closely with our research team and product engineers to efficiently deliver our product suite in a fast, cost-effective, and reliable manner.Construct robust inference infrastructure and monitoring systems for our product offerings.Enjoy substantial autonomy in shaping our products and directly influencing how cutting-edge AI is integrated across diverse devices and applications.What You BringAt Cartesia, we prioritize strong engineering skills due to the complexity and scale of the challenges we tackle.Proficient engineering skills with a comfort level in navigating intricate codebases, and a commitment to producing clean, maintainable code.Experience in developing large-scale distributed systems with strict performance, reliability, and observability requirements.Proven technical leadership, capable of executing and delivering results from zero to one amidst uncertainty.A background in or experience with inference pipelines, machine learning, and generative models.

Join the Sora Team at OpenAIThe Sora team is at the forefront of developing multimodal capabilities within OpenAI’s foundational models. We are a dynamic blend of research and product development, committed to integrating sophisticated multimodal functionalities into our AI offerings. Our focus is on delivering solutions that are not only reliable and intuitive but also resonate with our mission to foster broad societal benefits.Your Role as Inference Technical LeadWe are seeking a talented GPU Inference Engineer to enhance the model serving efficiency for Sora. This pivotal position will empower you to spearhead initiatives aimed at optimizing inference performance and scalability. You will collaborate closely with our researchers to design and develop models that are optimized for inference, directly contributing to the success of our projects.Your contributions will be vital in advancing the team’s overarching objectives, allowing leadership to concentrate on high-impact initiatives by establishing a robust technical foundation.Key Responsibilities:Enhance model serving, inference performance, and overall system efficiency through focused engineering efforts.Implement optimizations targeting kernel and data movement to boost system throughput and reliability.Collaborate with research and product teams to ensure our models operate effectively at scale.Design, construct, and refine essential serving infrastructure to meet Sora’s growth and reliability demands.You Will Excel in This Role If You:Possess deep knowledge in model performance optimization, particularly at the inference level.Have a strong foundation in kernel-level systems, data movement, and low-level performance tuning.Are passionate about scaling high-performing AI systems that address real-world, multimodal challenges.Thrive in ambiguous situations, setting technical direction, and driving complex projects to fruition.This role is based in San Francisco, CA. We follow a hybrid work model requiring 3 in-office days per week and offer relocation assistance to new hires.

ABOUT BASETENAt Baseten, we empower the leading AI companies of today, including Cursor, Notion, OpenEvidence, Abridge, Clay, Gamma, and Writer, by providing essential inference capabilities. Our unique blend of applied AI research, adaptable infrastructure, and intuitive developer tools enables innovators at the cutting edge of AI to seamlessly transition advanced models into production. With our recent success in securing a $300M Series E funding round, backed by notable investors such as BOND, IVP, Spark Capital, Greylock, and Conviction, we're on an exciting growth trajectory. Join our team and contribute to the platform that engineers rely on to launch AI-driven products.THE ROLEAs an Applied AI Inference Engineer at Baseten, you'll collaborate closely with clients to design, develop, and implement high-performance AI applications using our platform. You will guide customers through the entire process, from initial concept to deployment, transforming vague business objectives into dependable, observable solutions that meet defined quality, latency, and cost metrics.This position is ideal for innovative engineers eager to gain insight into how modern organizations scale AI adoption. You will thrive if you enjoy a multifaceted role that intersects product development, software engineering, performance optimization, and direct customer engagement.It’s essential to note that this position requires hands-on coding and software development, while also encompassing elements of product management, technical customer success, and pre-sales engineering.EXAMPLE INITIATIVESExplore insights from our Forward Deployed Engineering team through these blog posts: Forward Deployed Engineering on the frontier of AIThe fastest, most accurate Whisper transcriptionDeploy production-ready model servers from Docker imagesDeploy custom ComfyUI workflows as APIs...

At Inferact, we are dedicated to establishing vLLM as the premier AI inference engine, propelling advancements in AI by making inference both cost-effective and expeditious. Founded by the original creators and key maintainers of vLLM, we occupy a unique position at the convergence of models and hardware—an achievement that has taken years to realize.Role OverviewWe are seeking a talented Infrastructure Engineer to develop the distributed systems that facilitate inference on a global scale. In this role, you will design and implement essential layers that allow vLLM to deploy models across thousands of accelerators with minimal latency and maximum reliability. Our vision is to make deploying cutting-edge models at scale as simple as launching a serverless database. The complexities will be seamlessly integrated into the robust infrastructure you will be creating.

Anthropic is hiring a Research Engineer focused on Reinforcement Learning Infrastructure and Reliability. This role is based in San Francisco, CA. Role overview This position centers on building and maintaining systems essential to AI research. The work supports Anthropic’s reinforcement learning efforts, with an emphasis on infrastructure stability and performance. What you will do Collaborate with a team of specialists to develop and support key systems for AI research. Improve the reliability and efficiency of infrastructure supporting reinforcement learning projects. Apply technical expertise to advance Anthropic’s AI capabilities. Team environment Work alongside engineers and researchers dedicated to advancing AI reliability and performance. The team values collaboration and aims to enable new research while maintaining the stability of Anthropic’s core systems.

At Thinking Machines Lab, our mission is to empower humanity by advancing collaborative general intelligence. We're dedicated to crafting a future where everyone can harness the power of AI to meet their unique needs and aspirations.Our team comprises scientists, engineers, and innovators who have developed some of the most widely utilized AI products, including ChatGPT and Character.ai, as well as open-weight models like Mistral, in addition to renowned open-source projects such as PyTorch, OpenAI Gym, Fairseq, and Segment Anything.About the RoleWe are seeking a talented Infrastructure Research Engineer to architect and develop the foundational systems that facilitate the scalable and efficient training of large models using reinforcement learning.This position exists at the crossroads of research and large-scale systems engineering, requiring a professional who not only comprehends the algorithms behind reinforcement learning but also appreciates the practicalities of distributed training and inference at scale. You will have a diverse set of responsibilities, from optimizing rollout and reward pipelines to enhancing the reliability, observability, and orchestration of systems. Collaboration with researchers and infrastructure teams will be essential to ensure reinforcement learning is stable, rapid, and production-ready.Note: This is an evergreen role that we maintain on an ongoing basis to express interest. Due to the high volume of applications we receive, there may not always be an immediate position that aligns perfectly with your skills and experience. We encourage you to apply, as we continuously review applications and reach out to candidates when new opportunities arise. You may reapply after gaining more experience, but please refrain from applying more than once every six months. Additionally, you may notice postings for specific roles that cater to unique project or team needs; in those circumstances, you are welcome to apply directly alongside this evergreen role.What You’ll DoDesign, implement, and optimize the infrastructure that supports large-scale reinforcement learning and post-training workloads.Enhance the reliability and scalability of the RL training pipeline, including distributed RL workloads and training throughput.Create shared monitoring and observability tools to ensure high uptime, debuggability, and reproducibility of RL systems.Work closely with researchers to translate algorithmic concepts into production-quality training pipelines.Develop evaluation and benchmarking infrastructure to assess model performance based on helpfulness, safety, and factual accuracy.Publish and disseminate insights through internal documentation, open-source libraries, or technical reports that contribute to the advancement of scalable AI infrastructure.

About Our TeamJoin the Inference team at OpenAI, where we leverage cutting-edge research and technology to deliver exceptional AI products to consumers, enterprises, and developers. Our mission is to empower users to harness the full potential of our advanced AI models, enabling unprecedented capabilities. We prioritize efficient and high-performance model inference while accelerating research advancements.About the RoleWe are seeking a passionate Software Engineer to optimize some of the world's largest and most sophisticated AI models for deployment in high-volume, low-latency, and highly available production and research environments.Key ResponsibilitiesCollaborate with machine learning researchers, engineers, and product managers to transition our latest technologies into production.Work closely with researchers to enable advanced research initiatives through innovative engineering solutions.Implement new techniques, tools, and architectures that enhance the performance, latency, throughput, and effectiveness of our model inference stack.Develop tools to identify bottlenecks and instability sources, designing and implementing solutions for priority issues.Optimize our code and Azure VM fleet to maximize every FLOP and GB of GPU RAM available.You Will Excel in This Role If You:Possess a solid understanding of modern machine learning architectures and an intuitive grasp of performance optimization strategies, especially for inference.Take ownership of problems end-to-end, demonstrating a willingness to acquire any necessary knowledge to achieve results.Bring at least 5 years of professional software engineering experience.Have or can quickly develop expertise in PyTorch, NVidia GPUs, and relevant optimization software stacks (such as NCCL, CUDA), along with HPC technologies like InfiniBand, MPI, and NVLink.Have experience in architecting, building, monitoring, and debugging production distributed systems, with bonus points for working on performance-critical systems.Have successfully rebuilt or significantly refactored production systems multiple times to accommodate rapid scaling.Are self-driven, enjoying the challenge of identifying and addressing the most critical problems.