Everything is a Graph, Including LLM Applications! (and that's handy)

This was my talk at PyCon 2024 where I talk about our graph approach to building AI applications, especially since the release of Haystack 2.0.

In this talk, I give examples of common language model use cases such as retrieval augmented generation, embedding documents into vector databases, and how they can all be represented as graphs. I also talk about how having branches and loops can be really useful in many use cases, especially to build 'agentic' AU pipelines.

Transcript

1. EVERYTHING IS A GRAPH, INCLUDING LLM APPLICATIONS! (and that’s handy...)
2. @TUANACELIK /IN/TUANACELIK Tuana Çelik Lead Developer Advocate TUANACELIK OSS AI FRAMEWORK HAYSTACK AMSTERDAM, NETHERLANDS
3. @TUANACELIK /IN/TUANACELIK IT’S A NEW ERA 💅
4. @TUANACELIK /IN/TUANACELIK IT’S ALL POSSIBLE... maybe?
5. @TUANACELIK /IN/TUANACELIK Question Answering Given the documents, answer the question. Documents: {{ documents}} Question: {{ question }} Summarization Summarize the following text. Text: {{ text}} Question Generation Given the following document, generate some questions Document: {{ document}} Question: Translation Translate the following text to French Text: {{ text }} SOMETHING CHANGED IN 2022 An AI applications behaviour - defined by an instruction (prompt)
6. @TUANACELIK /IN/TUANACELIK Retrieval Augmented Generation
7. @TUANACELIK /IN/TUANACELIK Retrieval Augmented Generation Question Answering Given the documents, answer the question. Documents: {{ documents}} Question: {{ question }}
8. @TUANACELIK /IN/TUANACELIK Retrieval Augmented Generation Question Answering Given the documents, answer the question. Documents: {{ documents}} Question: {{ question }} Question
9. @TUANACELIK /IN/TUANACELIK Question Answering Given the documents, answer the question. Documents: {{ documents}} Question: {{ question }} Question Retrieval Augmented Generation
10. @TUANACELIK /IN/TUANACELIK Question Answering Given the documents, answer the question. Documents: Question: {{ question }} Question Retrieval Augmented Generation
11. @TUANACELIK /IN/TUANACELIK Retrieval Augmented Generation Retrieval Genration Query embedding Prompt Response Query
12. @TUANACELIK /IN/TUANACELIK Embedding Text and Storing
13. @TUANACELIK /IN/TUANACELIK Embedding Text and Storing Clean
14. @TUANACELIK /IN/TUANACELIK Embedding Text and Storing Clean Split
15. @TUANACELIK /IN/TUANACELIK Embedding Text and Storing Clean Split Embed
16. @TUANACELIK /IN/TUANACELIK Embedding Text and Storing Clean Split Embed
17. @TUANACELIK /IN/TUANACELIK Is this... a graph?
18. @TUANACELIK /IN/TUANACELIK Is this... a graph?
19. @TUANACELIK /IN/TUANACELIK PDF Is this... a graph?
20. Component Component Pipeline @TUANACELIK /IN/TUANACELIK Haystack 1.x: DAG
21. @TUANACELIK /IN/TUANACELIK Pipeline Haystack 2.0: DAG 📚 Announced March 11, 2024 Component Component Component
22. @TUANACELIK /IN/TUANACELIK Pipeline Haystack 2.0: DAG 📚 Announced March 11, 2024 Component Component Component
23. Fully open-source framework built in Python for custom LLM applications Provides tools that developers need to build state-of-the-art NLP systems Building blocks: Pipelines & Components Component Component Pipeline @TUANACELIK /IN/TUANACELIK 📚 Announced March 11, 2024
24. @TUANACELIK /IN/TUANACELIK Input 1 Input 2 Output ✨COMPONENTS ✨
25. @TUANACELIK /IN/TUANACELIK ✨COMPONENTS ✨ Input 1 Input 2 Output
26. @TUANACELIK /IN/TUANACELIK 🇬🇧TRANSLATOR 🇫🇷 from_lang to_lang documents documents
27. 0.35 0.8 0.1 0.21 0.47 0.62 0.73 0.51 @TUANACELIK /IN/TUANACELIK Is Taylor Swift queen of the world? Embedder Retriever PIPELINES
28. @TUANACELIK /IN/TUANACELIK (AGENTIC) PIPELINES
29. @TUANACELIK /IN/TUANACELIK HAYSTACK COMPONENTS Generators Embedders Retrievers Rankers Converters PreProcessors Classifiers Evaluators ....
30. @TUANACELIK /IN/TUANACELIK HAYSTACK COMPONENTS
31. @TUANACELIK /IN/TUANACELIK TIME TO BUILD LinkContentFetcher HTMLToDocument PromptBuilder Generator Prompt Response Query URL
32. @TUANACELIK /IN/TUANACELIK WEB QA
33. @TUANACELIK /IN/TUANACELIK WEB QA
34. @TUANACELIK /IN/TUANACELIK WEB QA
35. @TUANACELIK /IN/TUANACELIK WEB QA
36. @TUANACELIK /IN/TUANACELIK WEB QA
37. @TUANACELIK /IN/TUANACELIK WEB QA
38. @TUANACELIK /IN/TUANACELIK WEB QA
39. @TUANACELIK /IN/TUANACELIK 🚀HACKER NEWS SUMMARIES 🚀
40. @TUANACELIK /IN/TUANACELIK 🚀HACKER NEWS SUMMARIES 🚀 Summaries HackerNewsFetcher PromptBuilder Generator top_k
41. @TUANACELIK /IN/TUANACELIK 🚀NOTION QA 🚀
42. OUTPUT VALIDATION Meeting Notes PromptBuilder Generator Output Validator Valid replies Invalid replies & Error Message @TUANACELIK /IN/TUANACELIK 📚 Tutorial
43. BUILD YOUR OWN @TUANACELIK /IN/TUANACELIK PIP INSTALL HAYSTACK-AI Pipeline
44. BUILD YOUR OWN @TUANACELIK /IN/TUANACELIK PIP INSTALL HAYSTACK-AI Pipeline
45. @TUANACELIK /IN/TUANACELIK @HAYSTACK_AI THANK YOU! 🗂️ DOCUMENTATION 📚 TUTORIALS 📍 July 5th, Amsterdam 🧡 Come find me for a friendship bracelet PIP INSTALL HAYSTACK-AI