3 days ago
Responsibilities
- Build product-facing tools for browsing environments, inspecting trajectories, reviewing task quality, and understanding model behavior.
- Develop vendor-facing workflows for creating, submitting, testing, and iterating on reinforcement learning environments and training data.
- Create dashboards and observability tools for environment quality, evaluation results, data collection progress, and pipeline health.
- Design backend services and APIs connecting task authoring, data collection, evaluation, QA/QC, and reinforcement learning infrastructure.
- Own user-facing and internal products end-to-end from design through deployment.
- Collaborate with research, operations, engineering, and go-to-market teams to ship systems quickly.
Requirements
- Require 3+ years of full-stack software engineering experience building and shipping production systems.
- Proficiency in Python and a modern web stack such as React, TypeScript, or Next.js.
- Experience building backend services, APIs, and databases that connect multiple system components.
- Experience building tools for data inspection, review workflows, quality assessment, dashboards, monitoring, or observability.
- Hands-on experience with cloud infrastructure, Docker, CI/CD pipelines, and production debugging.
- Ability to build intuitive tools for technical and non-technical users with strong product judgment and independent execution.
- Strong communication skills across research, engineering, and operations stakeholders.
- Familiarity with AWS, Kubernetes, Terraform, or Grafana is preferred.
- Background in data collection, labeling, annotation, or evaluation platforms is preferred.
Benefits
- Full medical, dental, and vision coverage for US employees.
- 401k, commuter benefits, paid time off, and a company-wide holiday break.
- Equinox membership.
- Visa sponsorship and relocation support are available for strong candidates moving to the US or Singapore.
- Offices are located in San Francisco, California, and Singapore; remote candidates are welcome with 70–80% time-zone overlap with one location.