Staff Machine Learning Engineer

Unity Technologies•Mountain View, CA

45d•$172,200 - $283,900•Remote

About The Position

Unity is building next-generation intelligence systems for gaming: systems that help AI-native teams understand the market and get the most out of the frontier models. We are hiring a Staff Machine Learning Engineer to lead this effort. You will own the end-to-end architecture, build pipelines, systems, and models, and scale a cross-disciplinary team of engine developers, game developers, and machine learning engineers. The work spans market forecasting, outcome prediction, multimodal content understanding, evaluation, and agentic systems.

Requirements

Experience in machine learning with a track record of owning production ML systems end to end, from data to serving to measured business outcome.
Deep experience in performance marketing or ad-tech modeling: predicting acquisition cost, lifetime value, return on ad spend, retention or comparable outcomes.
Strong grounding in probabilistic prediction and calibration: quantile or interval prediction, coverage, bias correction and backtesting of forecasts.
Experience building LLM-based agent systems in production: orchestration, evaluation harnesses, prompt and model versioning, and cost control.
Strong Python and modern ML tooling, plus solid data systems fundamentals across a relational store, a columnar analytics store and an event stream.
Proven ability to drive technical direction across teams, mentor senior engineers, and influence decisions without formal authority.
Comfort building in a zero-to-one environment.
Sufficient knowledge of English to have professional verbal and written exchanges in this language since the performance of the duties related to this position requires frequent and regular communication with colleagues and partners located worldwide and whose common language is English.

Nice To Haves

Experience in mobile game user acquisition, LiveOps or monetization, on the studio or publisher side.
Experience with Unity or another game engine, or with automated and bot-driven playtesting.
Time-series and demand forecasting experience over noisy, multi-source market signals.
Experience with the Claude Agent SDK, Codex or similar frameworks for multi-agent software production.
Hands-on experience with BigQuery, ClickHouse, Kafka, Postgres, pgvector or their equivalents.
Publications or open-source contributions in forecasting, causal inference, calibration or agentic systems.

Responsibilities

Set the technical vision and architecture for our ML platform, from data ingestion, semantic layer and feature stores through model registry, evaluation and serving.
Make the calls that keep every model output reproducible and auditable: versioned data snapshots, model and prompt versions, and cost tracked per decision.
Define, lead, and spearhead engineering practices for the agentic era, including agentic development workflows and evaluation methodologies.
Mentor engineers from different backgrounds.
Partner with different functional teams and disciplines to translate expert human judgment into training signals.
Shape the roadmap with the product team.
Build predictors of commercial outcomes such as user acquisition cost, lifetime value and return on ad spend from large, noisy, multi-source signals.
Build forecasting models over market signals that detect emerging demand and estimate when a window opens, peaks and closes.
Deliver every prediction as a calibrated range with stated confidence, and own the ranking and portfolio logic that turns predictions into recommendations within a budget.
Design online experimentation with guardrails, safe exploration and clear human decision points.
Build evaluation harnesses for LLM-based agents, including automated playtesting and multimodal assessment of generated content.
Orchestrate frontier-model agents in production with strict data boundaries and cost control.
Own the feedback loop that scores predictions against realized outcomes and corrects bias, coverage and signal weighting over time.
Make the system report its own error in plain language so that creators and internal stakeholders can trust it.