Docugami has been using language models for multiple years in their mission to transform documents into data, initially starting with smaller models for text completion and OCR correction. As the models grew in size and complexity, they continued to invest in this space with question answering and Retrieval Augmented Generation (RAG) using their unique approach with a Document XML Knowledge Graph. They chose to host the language models in their cloud to ensure customer data confidentiality and started using LangChain early on, impressed by its expressive API and community. Docugami recently shared their learnings from an educational webinar hosted by LangChain, covering real-world challenges they've encountered with LLMs in production and how they're using LangChain, especially the new LangSmith tool, in their LLMOps flow. They discussed structurally chunking documents, documents as Knowledge Graphs, building complex chains with the LangChain Expression Language, debugging complex chain failures in production, and their end-to-end LLM Ops with LangChain + LangSmith.