Inference at Genesis (Paris, France)

Genesis· Paris, France·15h ago

Role details

Work type

Hybrid

Employment

Full-Time

Job description

Build low-latency inference pipelines for on-device deployment, enabling real-time next-token and diffusion-based control loops in robotics
Design and optimize distributed inference systems on GPU clusters, pushing throughput with large-batch serving and efficient resource utilization
Implement efficient low-level code (CUDA, Triton, custom kernels) and integrate it seamlessly into high-level frameworks
Optimize workloads for both throughput (batching, scheduling, quantization) and latency (caching, memory management, graph compilation)
Develop monitoring and debugging tools to guarantee reliability, determinism, and rapid diagnosis of regressions across both stacks

Deep experience in distributed systems, ML infrastructure, or high-performance serving (8+ years)
Production-grade expertise in Python, with strong background in systems languages (C++/Rust/Go)
Low-level performance mastery: CUDA, Triton, kernel optimization, quantization, memory and compute scheduling
Proven track record scaling inference workloads in both throughput-oriented cluster environments and latency-critical on-device deployments
System-level mindset with a history of tuning hardware–software interactions for maximum efficiency, throughput, and responsiveness

Culture: Values include candor, responsibility, limitless ambition, and a joyful journey. The team is described as multicultural, optimistic, and pragmatic.
Work model: Hybrid/office-based presence in Paris, San Francisco Bay Area, and London. No explicit remote policy mentioned, but the company emphasizes in-person collaboration.
Engineering culture: Builders get to work on the hardest problems in robotics and AI – from foundation models to hardware design to data systems. The team includes pioneers of generative simulation, Diffusion Policy, and GPU compilers.
Notable perks: The chance to shape a category-defining general-purpose robot; close collaboration with world-class investors and advisors; fast-growing startup with significant resources ($105M seed).

One-liner: Genesis AI is a full-stack robotics company building general-purpose robots with human-level dexterity and intelligence, powered by its own foundation model and hardware.
Entity Type: Private (Seed stage – $105M raised)
Headquarters: Paris, France (with offices in the San Francisco Bay Area and London)
Founded: 2024 (emerged from stealth in July 2025)
Founders: Zhou Xian (CEO), Théophile Gervet (President & Co-founder), Yi-Ling Qiao (Co-founder), Tsun-Hsuan Wang (Co-founder)

Primary industry: Robotics Engineering / General-Purpose Robotics
Target customers: Enterprises in manufacturing, laboratories, hospitals, and logistics; future expansion into homes.
Mission: “Build the world’s most capable robots that unlock new possibilities for humans.”

Eno (General-Purpose Robot): A full-body robot designed to operate reliably in factories, labs, hospitals, and homes. Driven by the GENE foundation model, it features 20 active, back-drivable degrees of freedom in its hands – matching human size and dexterity.
GENE-26.5 (Foundation Model): The AI model that powers Eno. It understands goals, reasons through changing conditions, and completes tasks end-to-end. Trained on massive human-based internet video data and proprietary simulation.
Data Collection Glove: A lightweight, sensor-loaded glove worn by humans to collect high-quality manipulation data in real-world settings (e.g., lab technicians, manufacturing workers). This glove bridges the “embodiment gap” and feeds the model with human skill data.
Simulation System: A proprietary simulation environment that accelerates model training and evaluation, reducing the iteration bottleneck.

Valuation: Not disclosed
Key Metric: Total funding of $105M (Seed round, July 2025)
Notable Investors/Partners: Eclipse (co-lead), Khosla Ventures (co-lead), Bpifrance, HSG; individual investors Eric Schmidt, Xavier Niel, Daniela Rus, Vladlen Koltun. Hardware manufacturing partner: Wuji Tech (China).
Growth Signals:
- Headcount grew from stealth to ~53–60 employees across Paris, Bay Area, and London (monthly growth +16% on LinkedIn).
- 29 active job postings as of mid-2026, with a +262.5% monthly increase in openings.
- Demonstrated a wide range of manipulation tasks (cooking, piano, Rubik’s cube, lab work) in public demos.
- In talks with multiple customers for commercial deployment in pharma and manufacturing.

Full-stack integration: Genesis controls both the AI model and the robotic hardware, enabling tighter optimization and faster iteration than companies that only do software or hardware.
Human-like dexterity: The robotic hand matches human size, shape, and 20 degrees of freedom, allowing direct transfer of human demonstration data without an “embodiment gap.”
Data flywheel: The combination of a low-friction data collection glove, simulation, and egocentric video creates a scalable pipeline for building a “human skill library.”
World-class team: Founders and early engineers have backgrounds from Mistral AI, NVIDIA, Google DeepMind, Apple Intelligence, Unity, Epic, and leading robotics labs (Berkeley, CMU).

Go full stack: Continue integrating model, hardware, and data collection to deliver a complete general-purpose robot.
Commercial rollout: Target early customers in pharmaceutical labs, manufacturing, and other industrial settings where dexterous manipulation is critical.
Scale data collection: Expand the use of the data glove (both internally and with third-party partners) to rapidly grow the model’s skill repertoire.
Global hiring: Aggressively grow the team across Paris, Bay Area, and London, especially in ML, simulation, rendering, and robotics engineering.

Sources