When people think of network science, they may picture social networks, disease outbreaks, or power grids. Few picture pet food. Yet for Guangyuan Weng, a Computer Science PhD candidate working with Professor Esteban Moro in NetSI’s Social Urban Networks Lab, a summer at online pet retailer Chewy showed how far network thinking can reach.
Guangyuan’s research sits at the intersection of machine learning and computational social science, studying what shapes access to opportunity in cities, and how large language models and recommender systems influence which places people see. As a Research Scientist II Intern on Chewy’s Marketing Science and Operations team, he was asked a different question, though. Which elements of the company’s large library marketing videos, distributed across platforms like TikTok, Facebook, and YouTube are associated with better perfomance? The fit was closer than it seems, because the role meant turning massive unstructured data into insight with multimodal AI.
“Chewy already had data on how well each video performed,” he explains, “but understanding why certain videos performed better than others was largely based on experience and intuition. My job was to build a reusable pipeline that could make this analysis more systematic.”
He built a multimodal LLM system that reads video, on-screen text, speech, and audio and converts them into structured creative elements, plus a human-in-the-loop workflow so the team could discover, refine, and consistently apply those elements. The system allowed the team to explore different creative questions and run tailored comparisons based on their business questions. For example: “We found that, for existing customers, videos emphasizing credibility tended to perform better than those focusing primarily on promotions.”
The project drew on his training in computer vision and LLM interpretability, but also on a network scientist’s way of seeing systems. Companies like Chewy run on networks, from supply chains to customer behavior to data flowing between teams. “In a large company, different teams own different products, data, and expertise,” Guangyuan says. “It’s almost like navigating a graph: you need to figure out who has the information you need, how to reach them, and how different pieces fit together. I also saw my project as one node in this network, connecting data, AI capabilities, and marketing decisions.” Navigating that graph also changed how he worked. “You can spend weeks building something technically interesting, only to realize it’s not what stakeholders actually need,” he notes. “Communicating early, getting feedback frequently, and keeping everyone aligned are just as important as the technical work.”
The work outlived the internship. A session Guangyuan had arranged with Chewy’s Chief Brand Officer was moved past his final week, so his team presented it to her and marketing leadership after he returned to school. “In academia, we measure impact through publications and citations,” he says. “It was rewarding to see another kind of impact, where something I built had the potential to inform real business decisions.”
Now finishing his PhD, Guangyuan encourages fellow students to cast a wide net. “Think less about what you’ve already done and more about what you’d like to contribute to,” he says. “Industry leaders value the critical thinking and intellectual independence that come with PhD training, even when the research topic isn’t an obvious match.” His summer is proof that networks show up wherever people, information, and decisions connect, even in pet care
