Hark

Build frontier user experiences for multimodal AI hardware, turning early design concepts into polished, shippable software. This individual contributor role bridges UI frameworks, embedded systems, and hardware to make new interaction models feel natural.

Build the low-level kernels and runtime systems that make Hark’s multimodal AI models run efficiently on constrained hardware. You’ll optimize transformer inference for latency, memory, and power across DSPs, NPUs, and emerging accelerators.

Lead the architecture and implementation of Hark’s on-device AI inference stack, from accelerator selection through custom kernels and runtimes. Build and manage the team delivering millisecond-latency transformer execution on battery-powered hardware.

Build and own Hark’s native iOS app, bringing real-time, multimodal AI interactions—including voice, text, vision, and streaming responses—to users’ devices. This high-ownership role combines polished product experiences with demanding performance and reliability goals.

Lead the operating system strategy and embedded software architecture for Hark’s next-generation consumer electronics portfolio. This principal-level role combines hands-on systems engineering, cross-functional technical leadership, and architecture ownership from early development through mass production.

Build Hark’s native Android app, bringing a real-time, multimodal AI agent to users through speech, text, vision, and persistent memory. This high-ownership role focuses on making the mobile experience fast, polished, and intuitive.

Join Hark’s Omni team to develop large-scale pretraining systems and multimodal foundation models spanning text, audio, and vision. You’ll improve model capabilities through better data, distributed training, architecture, and optimization.

Lead mid-training strategy for advanced multimodal and agentic models, improving reasoning, planning, tool use, and long-horizon decision-making. The role combines frontier ML research with distributed systems, synthetic data generation, reinforcement learning, and rigorous evaluation.