Machine Learning Engineer - Audio Specialist | Breaker

TLDR

• Build audio understanding models (STT, STS) from scratch for voice-controlled robotic systems
• Own the entire audio ML pipeline: data collection, training infrastructure, deployment, and optimization
• Deep audio/signal processing expertise required - this isn't general ML
• Travel for field testing in real-world conditions
• Join an exciting startup backed by globally recognised investors at the bleeding edge of physical AI

The way humans use robots is broken. Modern warfare demands more robots than we have operators. Every drone, ground vehicle, and maritime system requires dedicated training, manual control, and constant human oversight. One operator per robot. One pilot per mission. This operator bottleneck is the constraint on military capability today.

Breaker's AI agent, Avalon, breaks this constraint. Instead of piloting individual robots, operators command entire teams of autonomous systems across air, ground, and maritime domains, all through natural conversation.

A single operator can now coordinate multiple drones, ground vehicles, and other platforms simultaneously. Instead of flying search patterns on three different screens, you say “survey this area and flag anything unusual” - and a team of robots figures out how to divide the task, coordinate their movements, and report back what matters.

We're not just making robots easier to control. We're fundamentally changing the operator-to-robot ratio, turning small teams into force multipliers.

Our software deploys models directly onboard each robot, enabling real-time, intent-driven control even in contested environments with limited bandwidth. We're solving problems most AI companies never touch: sub-second inference on edge hardware with strict latency, power, and connectivity constraints.

We're backed by some of the best global investors and are growing our team across Austin, Texas, and Sydney, Australia. We're a small team of experienced engineers, moving fast on technology that will define how humans and machines work together for decades to come.

Join us if you want to help create the robots we were promised 🤖

Voice is the primary human machine interface our customers use with autonomous systems - and you'll own that capability from the ground up. You'll build speech recognition from scratch: building training pipelines, curating datasets, and shipping models that work reliably in real-world conditions where radio communication, wind noise, and varying audio quality are the norm, not the exception.

This is a rare opportunity to build foundational IP rather than integrate off-the-shelf solutions. You'll establish the entire ML infrastructure for the speech stack, from data collection strategies to model deployment on edge hardware. The work spans the full spectrum: running field tests to capture training data, experimenting with state-of-the-art architectures, and optimizing models to run efficiently on compute-constrained robotic platforms.

The technical challenge is unusually deep for an ML role. You're not just training models - you're making architectural decisions about how audio signals are processed, understanding the tradeoffs between different model families for varying audio quality scenarios, and building the instrumentation to track performance improvements over time. This is the kind of work that could result in patents and define how voice‑controlled robotics systems perform for years to come.

• Evaluate and implement state‑of‑the‑art architectures, making informed decisions about model selection based on audio quality constraints and deployment requirements
• Own metrics such as word error rate (WER), establishing baselines and demonstrating measurable improvements over time
• Build and maintain infrastructure for model training, including experiment tracking, performance monitoring, and version control
• Design data collection campaigns and field testing protocols to capture representative training data across varying environmental conditions
• Establish audio quality requirements and provide input on hardware selection for optimal model performance Deploy and optimize models for NVIDIA…