Why LLM Data Processing Pipelines Fail: UC Berkeley Research Insights | LangChain Interrupt

UC Berkeley PhD student Shreya Shankar shares research insights on why LLM data processing pipelines consistently fail in real-world applications. Based on systematic studies of developers building these systems, she reveals the core challenges: data understanding gaps and intent specification problems that cause the #1 complaint of 'this doesn't work.' Learn research-backed strategies to move beyond endless prompt iteration and build more reliable LLM pipelines for processing unstructured data.

Watch all of our recorded sessions from Interrupt here: https://interrupt.langchain.com/video/?utm_medium=social&utm_source=youtube&utm_campaign=q2-2025_interrupt-2025_co