Applied Research Intern, Proactive Intelligence & Customer World Models (PhD / Graduate Co-op)

About the Role

<p><strong>Team:</strong> Apollo — Block Applied R&amp;D<br><strong>Location:</strong> Remote (US / Canada)<br><strong>Duration:</strong> Fall/Winter 2026 co-op — 8 months, flexible start September 2026<br><strong>Level:</strong> Graduate student (MS or PhD, returning to your program after the co-op)</p> <h4><strong>About Apollo</strong></h4> <p>Apollo leads Block's efforts to build the <strong>Customer World Model (CWM)</strong>: a continuously evolving representation of each customer's goals, context, history, constraints, and likely future needs.</p> <p>The CWM powers <strong>proactive intelligence</strong> across Block's ecosystem. Instead of customers navigating products in search of features, intelligence observes their world, understands what matters, anticipates what comes next, and initiates actions on their behalf.</p> <p>We believe the next generation of AI products will not be defined by chat interfaces or isolated agents. They will be defined by rich world models that enable systems to reason over a customer's evolving state, make better decisions, and learn continuously from outcomes. Apollo designs, prototypes, and guides the development of this intelligence layer.</p> <h4><strong>About the role</strong></h4> <p>We're hiring a small cohort of graduate research interns to help build the foundations of proactive intelligence.</p> <p>This is not a traditional internship. You'll own a research problem end-to-end: framing the question, developing methods, running experiments, publishing findings, and, when successful, shipping your work into production systems used by millions of customers and sellers.</p> <p>You'll work at the intersection of representation learning, foundation models, reinforcement learning, causal reasoning, agentic systems, and product intelligence. The goal is not simply to build smarter models, but to build systems that develop a deeper understanding of customers and use that understanding to make better decisions over time.</p> <p>Past interns have shipped production systems within months and published their work in the same year.</p> <h4><strong>What you'll work on</strong></h4> <p>Depending on your interests and Apollo's roadmap, you'll focus on one or more of the following areas:</p> <p><strong>Customer World Models</strong></p> <p>Building rich representations of customers from event streams, financial activity, operational signals, and behavioral data.</p> <p>Examples include:</p> <ul> <li>Representation learning over long-horizon customer histories</li> <li>Event-based foundation models</li> <li>Multi-modal customer representations spanning structured, sequential, and graph data</li> <li>Memory architectures for long-term customer understanding</li> </ul> <p><strong>Proactive Intelligence</strong></p> <p>Developing systems that can anticipate customer needs and initiate helpful actions before being asked.</p> <p>Examples include:</p> <ul> <li>Opportunity detection and next-best-action systems</li> <li>Long-horizon planning and decision-making</li> <li>Preference and goal inference</li> <li>Learning when intervention creates value versus friction</li> </ul> <p><strong>Agentic Decision Systems</strong></p> <p>Building agents that reason over customer world models and take actions in real environments.</p> <p>Examples include:</p> <ul> <li>Tool use and planning</li> <li>Multi-step reasoning over customer state</li> <li>Autonomous workflow execution</li> <li>Recovery and adaptation under uncertainty</li> </ul> <p><strong>Learning from Feedback Loops</strong></p> <p>Developing methods that allow intelligence to improve continuously from real-world outcomes.</p> <p>Examples include:</p> <ul> <li>Reinforcement learning from customer and product feedback</li> <li>Reward modeling and preference learning</li> <li>Counterfactual evaluation</li> <li>Credit assignment over long decision horizons</li> </ul> <p><strong>Evaluation and Measurement</strong></p> <p>Building evaluation frameworks that predi