ML Engineer

About Alloy Robotics

We believe robotics is the industry of our lifetime, and data is its lifeblood.

Alloy is the AI-native data platform for robotics teams. Robot fleets generate enormous volumes of logs, video, telemetry, and mission data, and most of it ends up in silos where nobody can use it. Alloy turns it into one queryable layer with AI agents on top: engineers ask questions in natural language, get automated mission reports, and catch failures before they compound. The point is simple. Your robot breaks, you find out why, and your best people spend their time building the next robot instead of digging through the last one.

We've raised $16m, backed by Square Peg, Blackbird, and Airtree, with angels including senior leaders from OpenAI, Anthropic, Tesla, Waymo, and Halter. We partner with robotics teams across defence, agriculture, maritime, humanoids, construction, and medical, including Advanced Navigation, DroneForge, Breaker Industries, and Puralink. We're a lean team building from Sydney, and are adding our first US hires.

Where we are, and where we're heading

Alloy sits on data almost nobody has: real operations from real robot fleets, flowing through a product engineers use every day. Your job is the loop that turns that into compounding advantage: agents that get smarter with every fleet we serve, at a cost that keeps falling.

About The Role

You'll own the AI layer's learning loop: evaluations we trust, retrieval that works on messy fleet data, model and harness choices that bend the cost curve. This is applied ML shipped weekly, not a research seat.

What You'll Do

• Build the evaluation systems for quality, speed, and cost
• Make retrieval over fleet data state of the art
• Drive inference cost down: routing, caching, distillation
• Train task-specific models where APIs fall short
• Turn cross-fleet usage into datasets and detectors that improve with scale

You may be a good fit if you have

• 3+ years in applied ML
• Real LLM-systems experience: evals, retrieval, fine-tuning
• Python that ships product, not notebooks
• A pragmatic, measurement-driven style

Strong candidates may also have

• Models you trained yourself: embeddings, fine-tunes, task models
• Time-series or sensor-data experience
• Inference cost optimisation at scale
• Open-source work we can read

How We Work

We're building the company we always wanted to join: small, fast, and serious about the craft. The bar is work you'll be proud of in five years, shipped this week. Everyone owns outcomes end to end: the interesting problems don't come with instructions, and nobody will hand you a playbook. We question requirements, delete what shouldn't exist, and spend our energy where customers feel it. If you want clear lanes and a manager to set your week, we'll frustrate you. If you want real problems, real pace, and the best work of your career, you'll fit right in.

Compensation

• Competitive salary
• Additional compensation and benefits may include equity and flexible working arrangements

If this reads like you but you don't tick every box, apply anyway. We care more about trajectory than checklists.