Software Engineer, Infrastructure

About the role

ABOUT US

Tavus https://www.tavus.io/ is a research lab pioneering human computing. We’re building AI Humans: a new interface that closes the gap between people and machines, free from the friction of today’s systems. Our real-time human simulation models let machines see, hear, respond, and even look real—enabling meaningful, face-to-face conversations. AI Humans combine the emotional intelligence of humans with the reach and reliability of machines, making them capable, trusted agents available 24/7, in every language, on our terms.

Imagine a therapist anyone can afford. A personal trainer that adapts to your schedule. A fleet of medical assistants that can give every patient the attention they need. With Tavus, individuals, enterprises, and developers can all build AI Humans to connect, understand, and act with empathy at scale.

We’re a Series B company backed by world-class investors including Sequoia Capital, Y Combinator, and Scale Venture Partners.

Be part of shaping a future where humans and machines truly understand each other.

THE ROLE

We're hiring a Senior Software Engineer (Infrastructure) to own the systems behind CVI, our real-time conversational product. Every live conversation between a person and a PAL runs on infrastructure your team owns. You'll take goals like uptime, latency, and cost and chase them wherever they lead, including into backend services and product code.

WHAT YOU'LL OWN

- CVI's inference deployments. The GPU infrastructure serving live conversations across multiple providers and regions. You'll join as an early senior member of a growing infra team, working on projects like tuning the newest GPU generations and cutting cold-start and model load times so users wait less.

- Expanding our GPU footprint. You'll bring on new providers and regions, stand up clusters on EKS, and build the routing, scheduling, and throughput needed for fast weight loading.

- Uptime. You'll be one of the people pushing our uptime bar higher, along with the security and SOC2 work that keeps our infrastructure trustworthy.

- Fix what you find. When you see a problem, you have the trust and the mandate to fix it or flag it. Reworking our deploy pipeline so shipping is fast and boring is exactly the kind of thing you'd take on.

WHAT THIS ROLE HAS SHIPPED

- Multi-provider, multi-region inference infrastructure: routes live conversations across GPU providers and regions, so one provider's outage never becomes a user's problem

- CUDA optimizations for Phoenix, our video rendering model: doubled the frame rate by tracing and optimizing hot paths with our researchers

- Parallel conversations on a single GPU: several live conversations sharing one card, multiplying what the fleet can serve

WHO YOU ARE

- You own outcomes. You don't stop where "infrastructure" ends. If the fix lives in backend code or the CVI stack, you dive in, and you don't wait for a ticket to do it.