Machine Learning Engineer

What you will own

• Take speech and language models from evaluation into production, including the serving path and what it costs to run.
• Build evaluation sets that reflect the languages and accents customers actually speak, not the ones a public benchmark ships with.
• Instrument models in production: drift, latency, and the failure cases customers report.
• Work with the platform team on the data path, so that training data stays in the jurisdiction the customer requires.
• Document model behaviour and its limits to a standard a regulator can read.

What you bring

• You have deployed models to production and maintained them afterwards, not only trained them.
• You are comfortable with PyTorch or an equivalent framework, and with the serving stack around it.
• You understand speech or natural language processing well enough to know where the benchmarks mislead.
• You treat privacy and data residency as engineering constraints rather than as paperwork.
• You work in English. Other languages spoken in the group’s markets are an advantage.

Five steps, and you know where you stand at each one

1. 1ApplicationYou apply through this site. A person reads it — the first filter is not automated.
2. 2ConversationA call with the hiring manager about the work itself: what you have run, and what the role actually needs.
3. 3Work sampleA problem drawn from the real system, scoped to a reasonable session. It is discussed with you, never marked in silence.
4. 4PanelYou meet the people you would work with, and someone from outside the team who is there to ask the awkward question.
5. 5DecisionA yes or a no, with the reason. We tell you either way, and we tell you when to expect it.