Skip to content

README.md

Latest commit

 

History

History
42 lines (26 loc) · 2.23 KB

README.md

File metadata and controls

42 lines (26 loc) · 2.23 KB

Agentic RAG (Azure AI Search)

The code first creates a sample index, then demonstrates 3 sample Agentic RAG strategies to query it:

  1. Basic (No Reflection Included)
  2. Single Step Reflection
  3. Multi Step Reflection

Prerequisites

  • An Azure subscription, with access to Azure OpenAI.

  • An Azure AI Fondry Project

  • Bing Grounding to access public data on the web

  • Azure AI Search, any version, but make sure search service capacity is sufficient for the workload. We recommend Basic or higher for this demo.

  • A deployment of the text-embedding-3-large embedding model in your Azure OpenAI service. As a naming convention, we name deployments after the model name: "text-embedding-3-large".

  • A deployment of a at least one chat completion model in your Azure OpenAI service. You can try gpt-4o-mini for answer generation and query rewriting, and gpt-4o for evaluation and reflection.

  • Python (these instructions were tested with version 3.11.x)

We used Visual Studio Code with the Python extension to test this sample.

Setup Instructions

  1. Ensure your python environment is setup. You might want to setup a virtual environment
  2. Install requirements from requirements.txt: pip install -r requirements.txt
  3. Copy sample.env to .env and fill out all environment variables. You do not need to provide a Search key if you are using keyless authentication, but you must provide an Azure OpenAI key
  4. Run the ingestion script to setup your sample index: python ingest.py

Run the code

  1. Run the sample app to interact with the sample index: chainlit run app.py
  2. Navigate to http://localhost:8000 to chat
  3. You can load a sample conversation using the settings button
  4. You can select which Agentic RAG strategy to use to answer any question

Demo

Demo