Large language models (LLMs) can structure big data that either available or unavailable to researchers. I clearly distinguish two core approaches: retrieval and generation. Retrieval involves leveraging LLMs to extract structured information from extensive document corpora provided by user, while generation taps directly into the model's internal knowledge to produce structured outputs from minimal inputs. I will present papers that use these methods both separately: i) document-based retrieval techniques for extracting key data points and causal relationships from long and messy documents (e.g., www.causal.claims), alongside (ii) generative methods to construct granular production networks (e.g., aipnet.io) and comprehensive keyword dictionaries for political text classification (e.g., www.academicexpression.online). Additionally, I'll cover principles I've encountered regarding the LLM pipeline to effectively scale and adapt these tools to wide-set of research opportunities.