Staff Software Engineer, Machine Learning

About the role

Current is a leading consumer fintech platform transforming financial access for everyday Americans with over 6 million members. We provide access to financial solutions that seamlessly work together to solve the needs of our members and enable all Americans to build better financial futures. Based in NYC, our results-driven environment drives us to build better products, grow faster and empower everyone on our team to have an impact on our business and mission to improve financial outcomes. Current's Engineering team is dedicated to building our products and infrastructure. With our applications running on Google Cloud Kubernetes Engine, we support a proprietary banking core that can scale to handle millions of transactions a day. Our services run on MongoDB, with data exported to Google Cloud Storage and BigQuery for analytics. Batch feature computation runs on the JVM using Apache Beam and Dataflow, with dbt for the analytics estate and Airflow for orchestration. Our ML stack includes an in-house feature store and model registry with a serving proxy, and our data scientists develop models in Python. Machine learning drives decisions across the business: underwriting for liquidity products, fraud detection and risk exposure, marketing acquisition and spend. Models that ship faster and behave predictably in production are worth real money and real member trust. We are looking for a Staff Engineer, Machine Learning to join our Infrastructure team in New York.

Pay

This role has a salary range of $265,000 - $325,000.

Responsibilities

You will lead ML engineering initiatives across Current, with the goal of optimizing our model lifecycle: improving how we build, validate, deploy, and change models, and making that path faster and more repeatable as our model portfolio grows. This is a hands-on individual contributor role without direct reports, with room to grow into a team.

• Owning technical direction for the ML stack end to end: feature definition and computation, training data generation, training infrastructure, model serving, and production monitoring, along with the contracts between them
• Building tooling for training/serving consistency across analytics, batch computation, and live serving, accounting for differences in data sources and timing
• Designing how every deployed model stays linked to its dataset, feature versions, labels, and training code, to the standard model risk management expects
• Enabling data scientists to generate reproducible, point-in-time-correct datasets and run standard validation without an engineering ticket
• Setting the working contracts between the groups that build, consume, and govern models, and keeping the stack legible to people who don't read the code
• Measuring delivery time, engineering effort, and rework, and using that evidence to prioritize improvements
• Establishing a delivery baseline and proving the workflow on one production model with versioned features, a reproducible dataset, and reusable validation
• Extending those capabilities to additional models and measuring adoption and improvement against the baseline
• Standardizing model monitoring and defining production-readiness gates with Data Science, Risk, and service owners
• Evaluating build-versus-buy options for ML platform tooling against real production requirements