The Dagster team, inspired by the power of large language models like ChatGPT, decided to explore how they could use this technology to improve their own data orchestration solution. They wanted to create a Slack bot that could answer basic questions about Dagster using its vast knowledge base. The team started by investigating the feasibility of fine-tuning GPT-3 on Dagster documentation and eventually settled on using LangChain, an open-source library for building LLM-based applications. By leveraging LangChain's features, such as data augmentation and prompt engineering, they were able to construct a prompt that would elicit accurate answers from the bot. The team implemented this in a GitHub repository, where they used Dagster to manage their pipeline and cache the embeddings with Faiss. This allowed them to scale their application efficiently while maintaining high accuracy. With these changes, they created a Slack bot that could answer questions about Dagster's APIs and other technical concepts, demonstrating the potential of LLMs in data orchestration.