Member of Technical Staff, Pretraining Architecture and Scaling

About Hark Hark is an artificial intelligence company building advanced, personalized intelligence. One that is proactive, multimodal, and capable of interacting with the world through speech, text, vision, and persistent memory. We're pairing that intelligence with next-generation hardware to create a universal interface between humans and machines. While today's AI largely operates through chat boxes and decade-old devices, Hark is focused on what comes next: agentic systems that interact naturally with people and the real world. To get there, we're developing multimodal models and next-generation AI hardware together - designed from the ground up as a single, unified interface for a new era of intelligent systems. About the Role The Omni team at Hark is building the next generation of AI experiences beyond text, enabling models to understand and generate content across multiple modalities, including text, audio, and beyond. Our goal is to create seamless, real-time multimodal intelligence that powers intuitive and immersive user experiences. As part of the Omni team, you will build the multimodal foundation models at the core of this effort. You will work across the full pretraining stack, from large-scale data curation and mixture design to model architecture, scaling, distributed training, and evaluation, to create natively multimodal models that reason and respond fluidly across modalities. Your work will set the base capabilities that every downstream Omni experience is built on. You'll research architectures and training methods for Hark's multimodal foundation models. You'll design experiments, study how results change with model and dataset scale, and use the findings to improve model quality and training efficiency. The role includes implementing research ideas and developing training recipes with data researchers and training engineers. Responsibilities Conduct research on model architecture, optimization, and scaling to improve the capability and efficiency of our largest models. Establish baselines and run controlled experiments to evaluate which architectural and optimization changes improve performance at larger scales. Develop training recipes, including learning-rate schedules, context lengths, token budgets, and compute allocation. Investigate mixture-of-experts architectures, hybrid attention, and support for long-context training. Diagnose loss spikes, divergence, and numerical issues, working with training engineers to distinguish modeling problems from infrastructure failures. Requirements Hands-on experience training large models and making trade-offs among model quality, compute cost, and training stability. Strong experimental judgment, including designing controlled experiments that distinguish between competing hypotheses. Experience applying scaling laws to experiment design and evaluating extrapolations from smaller training runs. Experience implementing and debugging distributed training workloads across GPU clusters. Strong engineering skills, including implementing research ideas, profiling training workloads, and debugging model behavior. A record of improving model quality or training efficiency through research, engineering, open-source contributions, or production work. Bonus Qualifications Experience with mixture-of-experts routing, sparse architectures, or long-context methods. Work on data mixtures, curriculum, or tokenizer design. Kernel-level optimization or mixed-precision training experience. Experience with efficiency work aimed at constrained inference targets, including on-device. Compensation The US base salary range for this full-time position is between $180,000 - $450,000 annually. The pay offered for this position may vary based on several individual factors, including job-related knowledge, skills, and experience. The total compensation package may also include additional components/benefits depending on the specific role. This information will be shared if an employment offer is extended.