Member of Technical Staff — Developer Technology
Palo Alto, CA·Posted today
aipythongonodeawsllm
<div data-page-id="Tk0DdPSVaoiz1Jx8LC7jhatbpEc" data-lark-html-role="root" data-docx-has-block-data="false"> <h3 class="heading-3 ace-line old-record-id-Mo8id68lKoFoSwxC9Gtj9Bmrp6d">About the Role</h3> <div class="ace-line ace-line old-record-id-R3Fkd6rbtoEW4HxMjdOjYF25pHR">RadixArk is seeking a Member of Technical Staff, Developer Technology (DevTech) to make LLM inference and training dramatically faster, cheaper, and more accessible on modern GPU hardware. Our systems sit at the center of how modern AI is served and trained: SGLang is a high-performance inference engine that serves trillions of tokens daily across leading AI companies and research labs, and Miles is our reinforcement-learning post-training framework for large-scale LLM and MoE models. Your work directly advances our mission to democratize AI: every improvement you ship lowers the cost and raises the ceiling of what developers everywhere can build.</div> <div class="ace-line ace-line old-record-id-PGLXd4migowFtoxfOgqjbufmpEb"> </div> <div class="ace-line ace-line old-record-id-KqJfdkmypounenx4pYMjYLBVptV">As our technical face to a community of expert users and partners, you'll push the performance of SGLang and Miles through the lens of real production workloads. You'll profile and optimize GPU performance, enable new models and hardware, build kernels, deliver day-0 model support, and push the limits of inference and training. Working in close partnership with leading teams across the ecosystem, you'll turn their hardest, most ambiguous problems into concrete wins and clear guidance, and feed those improvements back into our systems and future roadmap.</div> <div class="ace-line ace-line old-record-id-YJLqd1l6FoTeHAxXqk4j5Ov9pcg"> </div> <h3 class="heading-3 ace-line old-record-id-SuCed5jcJooJI6xBq9sjwj4op1b">Key Responsibilities</h3> <ul class="list-bullet1"> <li class="ace-line ace-line old-record-id-GGuvdgL4ko999BxcGLjjmeC1pPd" data-list="bullet"> <div><strong>Accelerate AI workloads</strong><strong>.</strong> Profile and optimize GPU performance for real production workloads on current and next-generation hardware, root-causing bottlenecks from kernels to distributed multi-node systems.</div> </li> <li class="ace-line ace-line old-record-id-KLJIdoABaownJnxAh1ijZg45pgd" data-list="bullet"> <div><strong>Go deep in one or two focus areas.</strong> The team collectively covers the full stack; each engineer specializes in one or two tracks:</div> <ul class="list-bullet2"> <li class="ace-line ace-line old-record-id-C2HXd3510o7zsUxhfigjiBWPpof" data-list="bullet"> <div><em>Inference performance</em>: engine tuning, benchmarking, long-context and multi-turn optimization, parallelism strategy, production debugging</div> </li> <li class="ace-line ace-line old-record-id-LRZ9dtFHUo25ABx0NMqjDW8Wpvb" data-list="bullet"> <div><em>Kernels and model/hardware enablement</em>: custom CUDA/ROCm/Triton kernels, low-precision quantization, day-0 support for new models on new silicon</div> </li> <li class="ace-line ace-line old-record-id-DPoJdjFZtoee7Vxw4OxjwRxPpT9" data-list="bullet"> <div><em>Speculative decoding</em>: draft-model training, acceptance-rate tuning, cross-platform kernel adaptation</div> </li> <li class="ace-line ace-line old-record-id-GOlqdUTcToMQxTxm8hAjhxrup3c" data-list="bullet"> <div><em>Training systems</em>: RL post-training with Miles, FP8 training, elasticity, long-rollout and long-context efficiency</div> </li> </ul> </li> <li class="ace-line ace-line old-record-id-YaRrd5wbEo6QfpxoE1HjNkJXpqe" data-list="bullet"> <div><strong>Partner directly with the ecosystem. </strong>Turn ambiguous, high-stakes problems from expert engineers at our key partners into concrete wins, clear technical guidance, and reproducible cookbooks.</div> </li> <li class="ace-line ace-line old-record-id-UjHhd3aIjoEvtexcS1VjgIKlpXb" data-list="bullet"> <div><strong>Enhance SGLang and Miles</strong><strong>.</strong> Feed user-driven improvements back into our open-source systems and roadmap, so every win compounds across the ecosystem.</div> </li> </ul> <div class="ace-line ace-line old-record-id-SMDbdjp0IovRFtxZ4APjxUe4psg"> </div> <h3 class="heading-3 ace-line old-record-id-CVEtdUOxqoxXWqxxSspjN5qhpPQ">Qualifications</h3> <div class="ace-line ace-line old-record-id-Yxwcd4jNvoGSw6xau6FjoBIlppf"><strong>Minimum Requirements</strong></div> <ul class="list-bullet1"> <li class="ace-line ace-line old-record-id-QpvVd7n1OoYcIyxFgKHj4eP5pvd" data-list="bullet"> <div>4+ years of experience in GPU systems, LLM infrastructure, or performance engineering.</div> </li> <li class="ace-line ace-line old-record-id-HweidEu8DofOFtxia4ajyQgxpOg" data-list="bullet"> <div>Strong profiling and debugging skills: able to root-cause performance and correctness issues across the stack.</div> </li> <li class="ace-line ace-line old-record-id-GSUodVfoHoDCJZx1p7njRR5rpTe" data-list="bullet"> <div>Hands-on GPU programming experience in at least one of CUDA, ROCm, or Triton, and willingness to work across platforms.</div> </li> <li class="ace-line ace-line old-record-id-VrWPd6iEXoYrv6xI8KLj57jcpPe" data-list="bullet"> <div>Strong programming skills in Python plus C++ or CUDA.</div> </li> <li class="ace-line ace-line old-record-id-ELPjd8zJwouXYoxYxDrjCTP9pie" data-list="bullet"> <div>Comfortable making progress on hard, ambiguous problems with little context to start from, and fast to ramp into unfamiliar systems, codebases, and domains.</div> </li> <li class="ace-line ace-line old-record-id-ClVdd2rJJoD2y6xMNa2j7VR4pad" data-list="bullet"> <div>Ability to translate ambiguous asks into clear technical plans, verified cookbooks, and actionable recommendations, and to communicate credibly with expert engineering audiences.</div> </li> </ul> <div class="ace-line ace-line old-record-id-M5NkdDhFDo7ahcx7p1Oj3PvPp6f"><strong>Preferred (Bonus) Qualifications</strong></div> <ul class="list-bullet1"> <li class="ace-line ace-line old-record-id-FS0xdASDboy765xJsObj5KSgpec" data-list="bullet"> <div>Deep familiarity with LLM inference internals: distributed serving, parallelism, routing, KV-cache management, scheduling.</div> </li> <li class="ace-line ace-line old-record-id-ZhOQd6NlFogCXHxCxXdj7Wk1pAb" data-list="bullet"> <div>Experience with low-precision quantization and inference/training (FP8, INT8/INT4; NVFP4 or MXFP4 a strong plus).</div> </li> <li class="ace-line ace-line old-record-id-M89HdkiA7o9Y82xPJE2jzljfpXe" data-list="bullet"> <div>Experience writing and optimizing custom GPU kernels.</div> </li> <li class="ace-line ace-line old-record-id-T4LedRt0rowbDtx494MjkyXHpJh" data-list="bullet"> <div>Practical familiarity with speculative decoding methods such as Eagle, DFlash, or DSpark.</div> </li> <li class="ace-line ace-line old-record-id-KJQjdAwYMoEtT4xIvZxj0unLpVc" data-list="bullet"> <div>Working knowledge of large-scale distributed training: pre-training, SFT, RL post-training, elasticity, long-context workloads.</div> </li> <li class="ace-line ace-line old-record-id-WJ0gdIbeYoNan2xLWkSjcPAqpdd" data-list="bullet"> <div>Experience optimizing across both NVIDIA and AMD platforms.</div> </li> <li class="ace-line ace-line old-record-id-WTk7dEievoZxKBxc1xsjiGz5pOf" data-list="bullet"> <div>Hands-on experience with SGLang, Miles, vLLM, TensorRT-LLM, Megatron, or comparable frameworks; contributions to open-source AI/ML projects.</div> </li> </ul> <div class="ace-line ace-line old-record-id-Vd8OdL920okUvmxyQ1Mjps5Ypnh"> </div> <div class="ace-line ace-line old-record-id-Vd8OdL920okUvmxyQ1Mjps5Ypnh"> <h3 data-start="2687" data-end="2708"><strong data-start="2690" data-end="2708">About RadixArk</strong></h3> <p>RadixArk is an infrastructure-first company built by engineers who've shipped production AI systems, created SGLang (20K+ GitHub stars, the fastest open LLM serving engine), and developed Miles (our large-scale RL framework). We're on a mission to democratize frontier-level AI infrastructure by building world-class open systems for inference and training. Our team has optimized kernels serving billions of tokens daily, designed distributed training systems coordinating 10,000+ GPUs, and contributed to infrastructure that powers leading AI companies and research labs. We're backed by well-known infrastructure investors and partner with Nvidia, Google, AWS, and frontier AI labs.</p> <p>Join us in building infrastructure that gives real leverage back to the AI community.</p> <h3 data-start="3265" data-end="3284"><strong data-start="3268" data-end="3284">Compensation</strong></h3> <p data-start="3286" data-end="3456">We offer competitive compensation with meaningful equity, comprehensive benefits, and flexible work arrangements. Compensation depends on location, experience, and level.</p> <h3 data-start="3463" data-end="3487"><strong data-start="3466" data-end="3487">Equal Opportunity</strong></h3> <p data-start="3489" data-end="3576">RadixArk is an Equal Opportunity Employer and is proud to offer equal employment opportunity to everyone regardless of race, color, ancestry, religion, sex, national origin, sexual orientation, age, citizenship, marital status, disability, gender identity, veteran status, and more.</p> </div> </div>