Software Engineer, Early Careers focused on AI and UI

Roku•San Jose, CA

11d•Hybrid

About The Position

Roku is seeking early career Software Engineers focused on AI and UI to join their OS Platform UI team. This role is part of Project Ahsoka, a program for early career engineers building agentic systems. The team is responsible for Roku's home screen, Roku Channel, and other content discovery experiences that are used by hundreds of millions of devices worldwide. Engineers on this team own features end-to-end, from the OS platform layer through Application and UI to cloud integration, and collaborate with product, design, and cross-functional engineering partners. The role involves building multi-step agents, detecting and fixing regressions, developing internal tooling, processing real-time telemetry at scale, and addressing agent problems that involve hardware. Working fluently with AI tools is an explicit expectation for this role.

Requirements

Up to 4/6 years of professional engineering experience since finishing studies.
Strong systems or infrastructure fundamentals.
Experience with real-time data, telemetry, distributed systems, platforms, or device work.
Experience building and shipping something agentic that other people use (work project or serious personal project, multi-step rather than a wrapper around a chat window).
Ability to clearly account for how you stay accountable for code you did not type (e.g., design-first, review discipline, your own gates).
A view on where agents are not good enough yet and why.
Existing right to work in the US. Visa sponsorship is not available for this role.

Responsibilities

Build multi-step agents that perform real work in production.
Detect regressions across services, bisect commits that caused them, and propose fixes.
Generate and review code for other engineers to merge.
Develop internal tooling that colleagues will adopt.
Create skills, harnesses, and services to improve team efficiency.
Process real-time telemetry and events at streaming scale, under CPU and memory constraints.
Address agent problems that involve hardware, such as physical feedback, device fleets, and signals that cannot be simulated on a laptop.