Member of Technical Staff - ML Infra

About the company
Our client is a Seattle AI research lab with $60M raised, building photorealistic, real-time AI avatars that actually understand emotion. Their 19-person team holds PhDs from MIT, UW, Oxford, CMU and Johns Hopkins, with industry experience from Apple, Meta, Amazon AGI and Discord. They are training foundation models from the ground up for full-duplex audiovisual conversation: a system that listens, speaks, reacts and interrupts like a real person. Small team, unsolved problems, real work.
The role
Today's conversational avatars take 2 to 5 seconds to respond. Natural conversation needs sub-500ms. Closing that 10x gap means rethinking the entire serving stack, and this role owns it.
What you'll do

• Own and build the serving stack for our multimodal AI workloads, optimizing for latency, throughput and cost
• Architect and manage systems for real-time, long-lived WebRTC connections that keep video and audio smooth
• Build and orchestrate robust data pipelines for large-scale offline processing, evaluation and training using frameworks like Dagster or Ray
• Configure, maintain and optimize our GPU clusters with Kubernetes and Terraform
• Develop CI/CD, evaluation and versioning systems that enable safe, zero-downtime model deployments and rapid iteration
• Work closely with ML researchers and product engineers to build the foundational infrastructure powering visual conversational AI

What we're looking for

• 2+ years of full-time experience building and maintaining production-level ML systems
• 0 to 1 experience building ML infra at a VC-backed startup (Series C or earlier), on LLM inference systems, or on multimodal systems for video, audio or multimedia models
• Extensive experience building data pipelines as distributed systems
• Proficiency in Python and either Rust or Go
• Strong practical experience with Kubernetes, Terraform and cloud platforms
• A track record of optimizing systems for latency, throughput and cost, plus the ability to own complex projects and debug distributed systems

Bonus points

• Real-time video or audio streaming experience (WebRTC, low-latency infrastructure)
• Breadth across inference infrastructure, streaming and data engineering rather than depth in one narrow stack

Tech stack
Kubernetes, Terraform, Python, Rust, Go, Dagster, Ray, Airflow, WebRTC, vLLM, Triton Inference Server, TensorRT
Compensation and benefits

• $250,000 to $450,000 base, depending on seniority and experience
• Relocation assistance is available
• Visa sponsorship is available, including new H-1B applications

Location
On-site in Seattle, WA, 5 days per week. Full-time. Two openings.