Everything is a Graph, Including LLM Applications - Simplifying NLP with Graph-Based Thinking

Explore how conceptualizing NLP applications and LLMs as graphs can simplify development and enhance customization in this 34-minute PyCon US talk. Discover the structure of common NLP applications like retrieval-augmented generation (RAG) and learn to represent each step as a node in a graph. Examine the incorporation of branches and loops into these applications, and delve into building customized tooling for NLP applications in Python. Follow along with working examples using Haystack's pipeline structure and custom component API, including a private Notion question-answering app and a Hacker News post summarizer. Gain valuable insights into thinking of NLP applications as directed multi-graphs and leveraging open-source frameworks like Haystack to create tailored tools for Python developers.