ABOUT KOG

Kog builds the fastest LLM inference engine on standard datacenter GPUs. Our Kog Inference Engine generates 3,000 output tokens

per second per request on a single 8× AMD MI300X node and 2,100 on an 8× NVIDIA H200 node (FP16, batch size 1, no speculative

decoding).

The hot path is a monokernel implemented with handwritten CUDA (with PTX inline assembly) on NVIDIA, and HIP (with CDNA ISA inline

assembly) on AMD.

We optimize at the low level with engine/kernel/model co-design, using reverse engineering to understand and exploit the details

of how the GPU hardware works at the micro level.

We are a team of 11 people, including 10 engineers and 5 PhDs.

Test it at playground.kog.ai. Read the technical details on the Kog Labs blog.

WHAT YOU WILL WORK ON

You will perform experiments to understand GPU internals, find creative solutions to accelerate critical computational sections

used in LLM inference, and write optimized GPU kernels accordingly. Then test, profile, and optimize again.

Contribute to our monokernel pipeline, the single persistent GPU program that covers the full decode pass from QKV projection

to LM head sampling, across AMD and NVIDIA architectures.

Work on low-level GPU optimization, including impossibly-fast grid synchronizations and inter-GPU collectives, and optimized

GEMV, GEMM, and attention kernels across batch sizes and context lengths, with the memory-bandwidth-bound batch-1 GEMV regime

as the primary target.

Build profiling infrastructure inside a monokernel, including custom instrumentation, device-timestamp frameworks, and

per-stage analysis to translate machine behavior into concrete engineering decisions.

Scale the stack to third-party MoE models such as DeepSeek v4 and Qwen 3 to push generation speed on the models that matter in

production today.

Contribute to building AI agents that will perform GPU Engineering research and kernel optimization autonomously, calibrated to

hardware target and workload, starting from the inference foundations we are building now.

WHAT WE LOOK FOR

You have written GPU kernels where performance was the central constraint. Showing the code is a requirement to move forward in

the process.

PyTorch custom ops are an acceptable starting point if the kernels show a genuine understanding of the hardware below the

framework level.

Stronger signals include inline PTX or CDNA ISA in public repositories, experience with latency-sensitive execution paths,

understanding of why MBU matters more than MFU at batch size 1, and a background in inference engine components.

A top engineering school or a PhD with concrete GPU work counts, even without industry experience.

WHAT WE OFFER

Direct access to AMD and NVIDIA datacenter GPUs from day one

A team where creativity and technical judgment carry weight and where the people closest to the problem shape the key decisions

Problems that sit on the critical path of model execution speed and that directly influence what the system can become

A remote-friendly working model, with one mandatory week per month in our Paris office. Travel and accommodation covered by the

company.

Compensation aligned with top technical profiles in the Paris GPU Engineering market, including equity

Role

GPU Engineer

Locations

Paris

Published

June 9

Deadline

August 16

Tags for this job

Similar Jobs

Site Reliability Engineer (SRE) - AI GPU Clusters

Scaleway

Paris•Posted: Sep 10•Deadline: Aug 16

OUR STORY: 🇪🇺 Join Scaleway and shape the sovereign cloud of tomorrow ! Since 1999, we have been designing secure, sustainable infrastructures aimed at supporting the most ambitious companies. Historically known for our dedicated servers (Dedibox), we made a strategic shift to cloud computing in 2015. Staying true to our principles of simplicity, flexibility, and technical excellence, we have become one of the leading players in Europe in the sector. With the rise of artificial intelligence, we have strengthened our commitment, supported by the Iliad Group, which is investing €3 billion to develop a serious, sovereign AI alternative to American and Asian giants. Every day, thanks to our fast-growing portfolio of cloud and AI products (bare metal, containerization, serverless, AI, etc.), Scaleway proudly serves thousands of customer across the private and public sector, from corporations like France Télévisions or Hachette Livre, to fast-growing startups like Photoroom and Biolevate, to institutions like the City of Copenhagen. 📍 Our offices are located in Paris, Lille, Toulouse, Rennes, Rouen, Bordeaux and Lyon. WHY WE NEED YOU? Our growth is driving us to strengthen our SRE team to support and scale our production environments. Your mission will be to build and maintain reliable, observable, and secure infrastructure in order to ensure optimal service availability for our customers around the world. #HPC #AI #GPU #CLUSTERS YOUR FUTURE TEAM We work in a collaborative and international environment where the diversity of Scalers, combined with a spirit of sharing, helps bring new projects to life every day, advancing our ambitions together. You will join a newly formed team dedicated to building and operating Scaleway’s future AI infrastructure. As part of this group, you will design, maintain, and scale core systems and observability tools, partner with product teams, and ensure the reliability and performance of AI services across Scaleway. YOUR DAILY ROUTINE - Build a large AI infrastructure with monitoring, diagnosis, and remediation of production incidents- Troubleshoot high-impact production issues in collaboration with other engineering teams - Participate in an on-call rotation to handle incidents and ensure service continuity - Implement and maintain observability solutions to monitor AI infrastructure and application health - Contribute to AI infrastructure lifecycle management across different environments and countries - Promote and apply best practices in terms of stability, resiliency, scalability, and security - Maintain clear technical documentation for tools and procedures - Contribute to system and tool evolution based on production feedback - Collaborate closely with development teams to ensure infrastructure readiness- Participate in team rituals and knowledge-sharing initiatives ABOUT YOU 🎯 SOFTSKILLS : - Proactive and solution-oriented mindset - Passion for automation and continuous improvement - Strong collaboration and communication skills - Ability to work independently and in a team - Willingness to mentor and share knowledge 💻 HARDSKILLS : - Experience with Python, Go or C++ - Strong scripting skills (Bash, Python) - Hands-on experience with Linux systems (Ubuntu/Debian) - Preferred hands-on experience with GPU & HPC infrastructure - Knowledge of networking (TCP/IP, DNS, BGP, load-balancing, IPv6, etc.) - Familiarity with monitoring and logging tools (Prometheus, Grafana, Elastic, etc.) - Comfortable with Infrastructure-as-Code (Ansible, Salt, AWX, etc.) - Experience managing relational databases (MariaDB) - Understanding of CI/CD pipelines (GitLab) - Comfortable with English (written and spoken) WHAT YOU WILL FIND AT SCALEWAY ++++ Hybrid work: We offer up to 3 days of remote work per week. Offices: Our offices are spacious, dynamic workspaces with bold design, conveniently located near public transport. Most of our offices feature outdoor spaces (terraces) and bike parking facilities. Dining: Our chef provides a healthy meal service at the headquarters, and breakfast is available across all our sites year-round. Scalers working from regional sites enjoy a Swile card for lunches. Well-being commitments: Whether it’s access to a gym, daycare places, or discounted services for caring services, Scaleway is committed to supporting Scalers in maintaining a balanced life. International environment: With dozens of nationalities, Scaleway offers a stimulating environment where English is as widely spoken as French. Career & Mobility: Our managers value internal mobility, and opportunities to transition to other entities within the Iliad Group are accessible to all Scalers. 🚀 Why join the Scaleway adventure ? ✔ A rich and diverse product offering: Scaleway offers over 100 public cloud products in IaaS, PaaS, and AI. ✔ A cutting-edge technical environment: Scaleway provides modern infrastructures, including high-performance bare metal servers, to tackle exciting technical challenges. ✔ Commitment to responsible cloud: Scaleway is dedicated to a more responsible cloud, with data centers powered solely by renewable energy since 2017, minimizing our ecological footprint and holding top-level certification. 🔜 THE NEXT STEPS … - Discovery call with a recruiter (30 min) - Technical interview to validate your expertise (1h) - Interview with the manager to understand your approach to the role (45 min) - Interview with the Head of the Tribe to deepen your discussions and assess your fit with the team (45 min) - HR interview to tour our offices and meet your future colleagues

GPU Cloud

Research Engineer

Kog

Paris•Posted: Jun 9•Deadline: Aug 16

About Kog Kog builds the fastest LLM inference engine on standard datacenter GPUs. Our Kog Inference Engine generates 3,000 output tokens per second per request on a single 8× AMD MI300X node and 2,100 on an 8× NVIDIA H200 node (FP16, batch size 1, no speculative decoding). We co-design the model architecture and the execution engine together. Our Laneformer model uses Delayed Tensor Parallelism (DTP), a novel architecture that restructures the Transformer dependency graph so inter-GPU communication overlaps with computation rather than blocking it. We pre-trained a 2B-parameter DTP model on 6T tokens on 256 H100 GPUs. We are a team of 11 people, including 10 engineers and 5 PhDs. Test it at playground.kog.ai. Read the technical details on the Kog Labs blog. What you will work on You will imagine, design, and run experiments to understand how architectural decisions propagate through inference behavior, morph existing open-weight models into architecture variants optimized for speed, and turn findings into measurable gains in generation speed and model quality. * Design new model architecture variants, including routing strategies, attention mechanisms, and MoE structure, with execution constraints as a first-order design input. * Extend the Laneformer thesis by exploring inference-aware architectural variants such as DTP, Ladder Residual, and PT-Transformer, and finding what compounds at scale. * Own the post-training pipeline across fine-tuning, evaluation methodology, and adaptation of existing open-weight models toward architecture variants optimized for inference speed. * Scale the stack to large MoE models such as DeepSeek v4 and Qwen 3, working through routing, expert parallelism, and communication patterns at inference time. * Write up findings as research papers, submit them to top venues, and present them at conferences. * Contribute to building AI agents that will perform architecture research and training experiments autonomously, starting from the research foundations we are building now. What we look for * You have designed or changed model architecture, where the structure itself was the object of the work. Showing that work, a paper, a repository, or a thesis, is a requirement to move forward. * You reason about model design and hardware together, tracing how communication structure and layer dependencies shape inference behavior, with fluency in Transformers and MoE deep enough to weigh trade-offs. * Stronger signals include inference-aware architectural variants such as DTP, Ladder Residual, or PT-Transformer, and post-training methods such as fine-tuning, preference optimization, or quantization, including at research scale. * A top engineering school or a PhD with concrete architecture work counts, even without industry experience. What we offer * Direct access to AMD and NVIDIA datacenter GPUs from day one * A team where creativity and technical judgment carry weight and where the people closest to the problem shape the key decisions * Problems that sit on the critical path of model execution speed and that directly influence what the system can become * A remote-friendly working model, with one mandatory week per month in our Paris office. Travel and accommodation covered by the company. * Compensation aligned with top AI research profiles, including equity

Data Engineer - Foundational

Harmattan AI

Paris•Posted: Mar 6•Deadline: Aug 16

ABOUT US Harmattan AI is a next-generation defense prime building autonomous and scalable defense systems. Following the close of a $200M Series B, valuing the company at $1.4 billion, we are expanding our teams and capabilities to deliver mission-critical systems to allied forces. Our work is guided by clear values: building technologies with real-world impact, pursuing excellence in everything we do, setting ambitious goals, and taking on the hardest technical challenges. We operate in a demanding environment where rigor, ownership, and execution are expected. ABOUT THE ROLE As a Data Engineer on the Foundational team, you will serve as the "plumber" for deep learning, building the massive, high-performance data infrastructure required to power our foundational models. Based in Paris, you will manage terabytes—and eventually petabytes—of raw, unstructured, and noisy video data (EO and IR). Your mission is to ensure our ML engineers spend their time designing architectures, not waiting for data loaders or wrangling corrupted files. RESPONSIBILITIES * Multi-Modal Ingestion Pipeline: Build ETL/ELT pipelines to extract, decode, and store raw Electro-Optical (EO) and Infrared (IR) video from field logs into highly optimised formats like WebDataset, TFRecords, or Parquet. * Sensor Synchronisation & Alignment: Develop algorithms to programmatically synchronise EO and IR frames temporally and spatially to provide paired inputs for model training. * High-Throughput Data Loading: Architect storage-to-GPU pipelines to ensure multi-node training clusters maintain >90% GPU utilisation without I/O bottlenecks. * Distributed Processing: Write and optimise distributed data processing jobs using tools like Apache Spark, Ray, or Apache Beam to process thousands of hours of tactical video logs. * Data Quality & Versioning: Implement automated quality checks to filter corrupted or blank frames and maintain 100% reproducible training runs through robust versioning and lineage tracking. * Infrastructure Evaluation: Assess and implement advanced storage solutions (e.g., MinIO, S3 tiering) to manage growing datasets while optimising for cost and latency. CANDIDATE REQUIREMENTS * Educational Background: A BS or MS in Computer Science, Software Engineering, or Distributed Systems is highly preferred. Deep knowledge of operating systems, networking, and parallel computing is essential. * Technical Experience: 5-6+ years of experience building and maintaining terabyte-scale pipelines for unstructured data (video, images, or point clouds). * Performance Optimisation: Proven track record of maximising multi-node GPU utilisation and optimising data loaders for frameworks like PyTorch or JAX. * Tooling Expertise: Strong command of distributed computing tools (Spark, Ray, Beam) and ML data versioning tools (DVC, Apache Iceberg, or Pachyderm). * Adaptability & Ownership: A systems-thinker who thrives in a fast-paced startup environment and views messy data as an engineering problem to be solved via automation. * Commitment: 100% dedication to Harmattan AI’s mission of providing a defensive edge to allied nations through ethical, high-impact technology We look forward to hearing how you can help shape the future of autonomous defense systems at Harmattan AI.

AISpark