2026 Summer AI/ML Engineer Intern

Dex is Cursor for everyday operations.

We’re reimagining the browser as an intelligent workspace—one that understands what you’re doing and helps you do it faster. Instead of static pages and disconnected apps, Dex adds a layer of context-aware AI that works with you in real time.

We’re a small, fast-moving team, with 7-figure backing from top-tier investors, solving one of the most fundamental problems in computing: how humans and computers work together.

Join us in pushing the boundaries of what's possible with LLMs and browser-native AI. You'll work on cutting-edge problems in agent systems, context handling, and tool use, while collaborating directly with our research team to bring novel approaches to production.

• Browser agent and tool-calling multi-agent systems.
• Evaluation frameworks for memory, efficiency, and accuracy.
• Memory and personalization layer for workflows

• Research or work experience with RL environments, LLMs, modern AI frameworks, and/or ML.
• Experience with prompt engineering strategies.
• Strong foundation in Python and TypeScript.

• Designing and creating tool-calling environments to evaluate and benchmark agent systems
• Agentic systems that predict and execute users’ next steps in complex workflows.
• Mapping user paths on real world software to API functionality and action trajectories.
• A searchable, self-updating memory store for continuously learning agents.
• A system to interpret DOM snapshots, mouse click events, and keyboard inputs to select browser actions.
• A context composer that feeds relevant info into LLM prompts based on user interactions, page content, and memory.

Existing approaches—like computer-use data and Model Context Protocols—still overlook a fundamental element: a deep understanding of how individuals actually use software.

Every person navigates their workday with unique mental models and personal systems for interacting with platforms and staying organized. These invisible frameworks shape productivity and workflow in ways that generic data can’t capture.

How can we systematically capture, structure, and teach these personal workflows to AI—enabling it to become a truly proactive, personalized extension of each user?