Job description

About Voltai Voltai is the leading AI company building agentic systems and frontier foundation models for semiconductor and electronics design. Backed by Sequoia Capital, we’re putting AI in the hands of hardware engineers in over 70% of the world’s largest semiconductor and electronics companies to have effortless control over their next-generation chip and board designs, powering the future of automotive, industrial automation, consumer electronics, IoT, and semiconductor manufacturing. About the Team Our founding team consists of IOI/IPhO olympiad medalists, Stanford professors, ex-CTO of Synopsys, and our business leadership has scaled revenue in their previous companies to over $1.5bn. At Voltai, we are combining the world’s best talent in the intersection of software and hardware. Key Responsibilities • Design and maintain high-performance ML pipelines for training, evaluation, and inference of LLMs and retrieval-augmented systems, with a focus on hardware efficiency and throughput • Optimize core transformer operations at the kernel level, designing and tuning custom kernels and low-level implementations for GPU-accelerated workloads • Implement and integrate low-precision computation techniques to reduce memory footprint and accelerate inference with minimal accuracy degradation • Build and maintain inference engines for on premises deployments • Architect distributed training and inference systems • Collaborate closely with researchers and infra teams to bring cutting-edge model innovations into production • Interface directly with enterprise hardware environments, tuning performance based on real-world deployment constraints Required Skill Sets • Languages: Expertise in C, C++, or Rust • Design and Optimize CUDA Kernels for LLMs : Develop and fine-tune custom CUDA kernels to accelerate core transformer operations • Implement Low-Precision Computation Techniques : Apply quantization methods like AWQ and GPTQ to reduce model size and inference latency. Ensure minimal accuracy loss while maximizing throughput on GPU architectures with familiarity with concepts like GGUF and GGML • Develop and Maintain High-Performance Inference Systems : Build, improve, and maintain inference engines such as vLLM, SGLang, and TensorRT with a focus on low-latency and high throughput • Architect Distributed Training and Inference Solutions : Design systems that support model parallelism (tensor, pipeline, expert etc) to enable efficient training and inference across multiple GPUs and nodes • Integrate Research into Production Systems : Translate cutting-edge research findings into robust, production-ready systems. Ensure that innovations in model architectures and optimization techniques are effectively deployed. • Monitor and Optimize System Performance : Implement monitoring tools to track system metrics, identify bottlenecks, and optimize performance Bonus Points • Some background in hardware/electronics, gained through professional, academic, or personal projects • Contributions to open-source initiatives • Notable awards or publications in leading journals/conferences • Experience thriving in a fast-paced, hyper-growth startup environment Our Benefits • Unlimited PTO: Recharge when you need it, no questions asked. • Comprehensive Health Coverage: Medical, dental, and vision insurance for you and your dependents. • Free Meals and Snacks : Daily lunches, dinners, and snacks in the office. • Professional Growth: We invest in your continuous learning and offer opportunities to expand your skills. • Visa Sponsorship : We welcome global talent and provide visa sponsorship to support qualified candidates.

Requirements

Must-haves

Design and maintain high-performance ML pipelines for training, evaluation, and inference of LLMs and retrieval-augmented systems, with a focus on hardware efficiency and throughput
Optimize core transformer operations at the kernel level, designing and tuning custom kernels and low-level implementations for GPU-accelerated workloads
Implement and integrate low-precision computation techniques to reduce memory footprint and accelerate inference with minimal accuracy degradation
Build and maintain inference engines for on premises deployments
Architect distributed training and inference systems
Collaborate closely with researchers and infra teams to bring cutting-edge model innovations into production
Interface directly with enterprise hardware environments, tuning performance based on real-world deployment constraints
Required Skill Sets
Languages: Expertise in C, C++, or Rust
Design and Optimize CUDA Kernels for LLMs : Develop and fine-tune custom CUDA kernels to accelerate core transformer operations
Develop and Maintain High-Performance Inference Systems : Build, improve, and maintain inference engines such as vLLM, SGLang, and TensorRT with a focus on low-latency and high throughput
Architect Distributed Training and Inference Solutions : Design systems that support model parallelism (tensor, pipeline, expert etc) to enable efficient training and inference across multiple GPUs and nodes
Monitor and Optimize System Performance : Implement monitoring tools to track system metrics, identify bottlenecks, and optimize performance

Nice-to-haves

Some background in hardware/electronics, gained through professional, academic, or personal projects
Contributions to open-source initiatives
Notable awards or publications in leading journals/conferences
Experience thriving in a fast-paced, hyper-growth startup environment

Application prep

Use the role’s language and application details while tailoring your materials.

Keywords to mirror

aimlrustdesign systems

Platform: gem
Effort: medium

Benefits

Benefits detected in the listing.

Dental

Vision

Pto