Senior Machine Learning Engineer - Active Learning

Senior Machine Learning Engineer - Active Learning

About Sense

Sense is a climate tech company that turns electric signals in homes and on the grid into real-time insight about energy use, devices, and faults. These insights help people lower their electric bills and use less electricity, and they help utilities prevent wildfires, among other applications. These insights come from load disaggregation and fault detection models that are running on meters today. The technology team at Sense researches, trains, deploys, and supports these models.

We are looking for an experienced Senior Machine Learning Engineer with expertise in active learning systems to help us scale our data operations so we can continue to improve our core production models and build the foundation model of electrical devices and energy.

The role

This Senior Machine Learning Engineer will own the loop from data to labels to improved models. You will use active learning and weak supervision methods to generate labels, decide what to label next, and turn errors we find in the field into new labels.

This is a hands-on role. You will be expected to write code, train models, analyze data, and document experiments. You will work with data scientists, annotators, product, and engineers to get labels and improve models. You will direct the development of annotation tools and the work of our annotators to speed up and scale labeling.

Success in this role is measured by the reduction in cost per annotation (i.e., time per annotation), while keeping label quality at or above our targets.

Responsibilities:

• Design and run active learning to choose which homes, devices, and time periods to label next, based on where models are uncertain or wrong.
• Turn bad detections, field accuracy results, and user corrections into new ground truth, and track whether retrained models improve.
• Find rare events and hard cases in home data, such as uncommon or mislabeled devices, and add them to labeling and evaluation sets.
• Own the ground truth plan: which data, how much, and in what order.
• Direct a team of annotators: set priorities, track throughput, and provide guidance. Set data quality standards.
• Direct development of our annotation tool. Add features that enhance our ability to collect new and better data, including features that rely on models.
• Measure and improve label quality through metrics, outlier detection, human review, etc.
• Work with data engineers and ML operations to version labels and track which labels trained which model. Keep evaluation sets separate from training data so offline results predict performance in the field.
• Be a good teammate: review code, provide peer feedback on research and experiments, and contribute to shared documentation. Help mentor/coach less experienced team members.