ML Engineer

Location: Santa Clara, CA, United States

Job Type: Full-time

Salary: $160K - $340K

Equity: 0.10% - 0.50%

Experience: 3+ years

Visa: Will sponsor

About Blue: Blue is building a new kind of phone, with your chief of staff built in. Blue sees what you do and reads every notification, text, and email for you. Based on everything it’s learned about you, it decides what to do with each one: handle it and leave you an FYI, or draft a reply and wait for your OK. Anything in your life is one ask away from being off your plate. Blue can manage complex long term tasks across apps

Delegation this deep isn’t possible from a single agent in an app. It takes owning the operating system, the whole intelligence stack, and, ultimately, the hardware. Today the Blue OS runs on Google Pixel hardware while our own hardware is in development. We’re Y Combinator-backed, seed-funded, going into our series A, and our first customers already use it as their daily phone.

It works from day one. When you switch, Blue reads everything that comes over from your iPhone, so it already knows the people in your life, how you write, and the passwords, cards, and IDs it needs, kept in a secure vault you both share. No onboarding, and no explaining yourself.

The role

Blue's background agent reads every notification, text and email and operates apps on your behalf. Today it runs mostly on cloud models; we are going to train our own, starting with the tasks Blue does most, and run them on our servers and eventually on the phone. You will also own the models that already run on the phone and the voice experience they power.

This role is open at mid and senior levels. Depending on your background, you will own some of the areas below.

What you'll own

• Agent models: Post-training open-weight vision-language models on our agent data, so Blue can read the screen, choose the right action and finish tasks, then moving tasks off cloud models one at a time as ours match them

• Evaluation: Measuring our models against cloud models on real tasks, with our evals team, before anything ships

• Cost and speed: Distillation, quantization and serving choices that make our models cheaper and faster than the cloud models they replace

• Voice and on-device: The hotword model, voice authentication, hands-free duplex voice, realtime latency, dictation quality, small models for notification triage

• Data anonymization: Working with privacy and security on what we record, and how it is redacted

• Review and mentoring: Reviewing ML changes from teammates and AI agents, and helping early-career engineers grow into strong ones

What we're looking for

• For agent models: you have fine-tuned or post-trained large models and shipped them to real users, not only to a benchmark, ideally vision-language models, tool use, or agents that act in a UI

• For voice and on-device: you have trained smaller models, or systems of several models working together, from scratch and deployed them, such as wake word, speech recognition or speaker verification, on runtimes like LiteRT, ONNX Runtime or llama.cpp

• You evaluate carefully and trust real-world measurements over benchmarks

• You do the arithmetic early: training cost, serving cost per user, latency

• You give real reviews, keep PRs clean, and have mentored junior engineers

• You work well with ambiguity and can move quickly without sacrificing quality

• You want to make a Blue Phone your daily phone

Nice to have

• Reinforcement learning for agents, or reward modeling

• Privacy-preserving training data practices

• Experience reviewing AI-generated code, or working with coding agents

Why join now

• Few companies have an agent running on real people's phones all day; the data and the problem are very unique

• We have very wide coverage over different types of problems we are solving for, and you get to pick what you want to work on.

• We don't skimp on compute and do not have a ton of broken infra that gets in your way