The Data Flywheel team closes the gap between benchmark performance and useful performance in the real world. We identify and build the signals, data, and feedback loops that turn model usage into rigorous evaluations, targeted training data, and measurable improvements in future generations of models.

This is a hands-on technical role at the intersection of research and deployment. You'll take ambiguous model behaviors from first observation through measurement, intervention, and validated improvement, working across evaluation, human and synthetic data, infrastructure, post-training, and live deployments. You'll collaborate closely with researchers and engineers across the company, as well as customers, partners, vendors, and the open-source community.

Identify high-value data sources and partnership opportunities, deeply understand the underlying use cases, and translate them into representative evaluations

Bring new data sources online, from initial partner conversations and data scoping through quality validation and integration into production evaluation and training pipelines, where appropriate

Design and build evaluations, graders, and feedback loops that make priority real-world model behaviors measurable

Analyze model performance and failure modes, then translate those insights into targeted datasets, reward signals, and training interventions

Develop human and synthetic data strategies for capabilities where existing data is insufficient, including designing and running evaluation and data-collection programs with vendors

Build the infrastructure and pipelines needed to ingest, inspect, version, and evaluate data reliably at scale

Collaborate closely with pre-training, post-training, applied, and partnership teams to turn new signals into measurable model improvements

Member of Technical Staff, Data Flywheel

Responsibilities

Qualification

Required

Benefits

Company

Reflection AI

More roles at Reflection AI