Hathora | NYC (In-Person) | $160-200k + equity

We're building the fastest voice AI inference platform in the world. Our low-latency models platform (models.hathora.dev) runs voice models globally to minimize network latency. As voice AI explodes (ElevenLabs, Cartesia, Hume raising billions), we're positioning to own the infrastructure layer through a marketplace of optimized models.

What you'll do

• Optimize inference latency and throughput for voice AI models using CUDA, kernel-level tuning, and engine modifications (vLLM, Nemo, etc)
• Deploy and scale models on our Docker/Kubernetes infrastructure with high reliability
• Own the performance stack end-to-end: from GPU kernels to production deployment
• Rapidly integrate new models as they're released and ensure we're the fastest option available
• Publish research and speak at conferences—establish yourself and Hathora as the authority on voice inference performance

What we're looking for

• Proven experience with GPU programming (CUDA) and low-level performance optimization
• Contributions to inference engines (SGLang, vLLM, etc.) or PyTorch internals preferred
• Comfort with deployment infrastructure (Docker, Kubernetes)
• Demonstrated curiosity and depth—whether through projects, contributions, or research
• Any experience level welcome if you've gotten your hands dirty and want to go deeper

Why join

• Build the inference stack from scratch as the voice AI market explodes—your decisions will define how the industry runs models
• Own the entire inference stack for a rapidly growing platform
• Solve hard technical problems: sub-100ms voice responses at global scale
• Backed by Upfront Ventures, Founders Fund, Lunar Ventures

About Hathora