Episode

Make Azure AI Real: Azure AI Cognitive Search + ChatGPT

with Matt Gotteiner, Korey Stegared-Pace

Make Azure AI Real

Dive deep into learning how to chat with your data. Explore how to use AI-powered search and understanding to facilitate natural language question and answering over your data sets. This session is oriented towards developers and enthusiasts interested in using large language models to comprehend natural language questions and respond effectively with relevant information from a dataset.

Chapters

  • 00:00 - Intro
  • 01:26 - What new skill are you learning?
  • 04:20 - Chat your Data with Azure OpenAI
  • 05:38 - Citations
  • 07:49 - Generative AI content creation by API
  • 08:35 - Retrieval Augmented Generation
  • 10:00 - How does the model handle questions it doesn't know the answer to?
  • 12:08 - What are major differentiators for CogSearch vs other vector stores?
  • 17:05 - Best practices if source data is not static
  • 19:03 - What temperature is recommended when doing RAG?
  • 22:22 - Can we connect to Azure Synapse and Delta Lake?
  • 23:26 - How do we generate follow up questions?
  • 28:24 - Embeddings
  • 31:07 - Document chunking
  • 35:21 - PDF parser
  • 36:39 - Will Azxure OpenAI be available like an other service in Azure soon?
  • 37:31 - Semantic Kernel/Langchain/Haystacks
  • 41:34 - How do these models see the world?
  • 45:52 - What is the best way to come up with a hard threshold for search scores?
  • 53:39 - Recommended architecture patterns for resilient application
  • 55:19 - Assistant API
  • 56:16 - Thought process tab

Developer
Azure