In the next five years, the world will undergo profound changes.

Machine learning models are elevating our daily experiences — from self-driving cars and vision tests to cancer detection, restoring sight to the visually impaired, enabling speech for those who cannot speak, and influencing our consumption and entertainment choices. These AI systems are already woven into the fabric of our lives and are poised to guide the trajectory of our future.

Imagine a world where we can create endless content, whether it’s immersive TV series featuring ourselves as protagonists or personalized tutors that ensure no student is overlooked. Upcoming advancements will allow us to augment our memories with customized foundation models, developed through Reinforcement Learning from Human Feedback (RLHF), and directly linked to our thoughts via Brain-Machine Interfaces. This will blur the lines between human cognition and machine intelligence, propelling us into an era of unprecedented human growth.

To support this vision, we need an expansive, globally accessible, and uncensorable computational infrastructure. Gensyn is pioneering a machine learning compute protocol that transforms computing resources into a continuous, commodity-like service that operates beyond centralized control, akin to electricity. This approach accelerates AI advancement while ensuring that the revolutionary power of this technology is available to everyone through a free market.

Our Core Values:

AUTONOMY

• No need for approvals – we foster a culture of constraints instead of permissions.
• Take ownership of your projects and establish goals and deadlines proactively rather than waiting for assignments.
• Engage with the context around your projects; proactively seek information and share your progress.
• We operate without middle management, empowering every individual to drive initiatives.

FOCUS

• Our compact teams are designed to minimize misalignment and bureaucracy, allowing for agility that outpaces larger traditional teams.
• Embrace simplicity in our designs and frameworks.
• Protect the company’s time by avoiding unproductive meetings and unnecessary distractions.

REJECT MEDIOCRITY

• Provide immediate, direct feedback rather than deferring discussions or allowing issues to fester.
• Adopt a mindset of continuous learning and growth, pushing beyond perceived limitations.

Key Responsibilities:

Lower deep learning graphs from popular frameworks (such as PyTorch, TensorFlow, Keras) to an intermediate representation (IR) for training, with a strong emphasis on reproducibility.

Develop innovative algorithms for transforming compute graph representations into various operator formats.

Own two of the following compiler domains:

• Front-end: Manage interactions between popular Deep Learning frameworks and Gensyn’s IR. Create ONNX transformation passes for middle-end optimization.
• Middle-end: Author compiler passes for training-based compute graphs, integrate reproducible Deep Learning kernels into code generation, and troubleshoot compilation processes.
• Back-end: Convert IR from the middle-end into optimized GPU machine code.

Essential Qualifications:

Fundamental understanding of compiler operations, including knowledge of traditional compilers (e.g., LLVM, GCC) and graph traversal necessary for their operation.

Strong software engineering background with practical experience in producing and delivering production-ready code.

Familiarity with parallel programming, particularly as it applies to GPU operations.

Open to learning Rust, as our organization utilizes it as the primary language for our codebase.

Experience working with:

• High-Level IR/Clang/LLVM through middle-end optimization; and/or
• Low-Level IR/LLVM with a focus on GPU-specific optimizations.

Self-driven, with strong communication capabilities, both verbal and written.

Able to thrive in a highly autonomous applied research environment.

Desirable Qualities:

Comprehensive understanding of computer architectures designed for training neural network graphs (Intel Xeon CPUs, GPUs, TPUs, or custom accelerators).

Experience with Rust in systems-level programming.

Contributions to open-source compiler stacks.

Strong grasp of compilation processes related to high-performance computing architectures (CPU, GPU, custom accelerators, or a combination of these).

Proven expertise in CPU and GPU architectures, numeric libraries, and modular software design.

Awareness of current trends in Deep Learning architectures and foundational knowledge in training methodologies, along with experience in machine learning frameworks such as PyTorch, TensorFlow, or scikit-learn.

Exposure to Deep Learning Compiler frameworks like TVM, MLIR, TensorComprehensions, Triton, and JAX.

Experience with developing and optimizing high-performance GPU kernels.

Note: We welcome applications from candidates who do not meet every listed criterion, as there may be opportunities at various experience levels.

Compensation & Perks:

Attractive salary package plus equity shares and access to token pools.

Work remotely from anywhere between the U.S. West Coast (PT) and Central Europe (CET) time zones.

Enjoy four fully-funded company retreats around the globe each year.

Access to the tools and equipment you need to succeed.

Paid sick leave for your peace of mind.

Comprehensive private health, vision, and dental insurance, extending to spouses and dependents.