akihiro-inui/promptflow-hybrid-search-rag-japanese

A working prototype of performing Semantic Hybrid Search (RAG) with PromptFlow and Azure AI Search

Best-in-class RAG with PromptFlow and Azure AI Hybrid Search

This repo demonstrates building a generic RAG pattern with PromptFlow and Azure AI Hybrid Search

Why PromptFlow ?

• To simplify development of complex LLM app workflow
• To systemtically evaluate and quantify LLM quality
• To simplify operationalizing LLM app as a docker on managed endpoints

Why Hybrid Search ?

• Azue AI Search makes it possible to combine the best of Keyword search (BM25), Semantic Search and Vectorized Search for an improved RAG performance (see benchamrking)

Instruction

1. Open .env.template file and configure following Azure service's connection details.

   Key	Value
   AZURE_SEARCH_SERVICE_NAME	Azure AI Search service name (e.g. jixjia-search-dev)
   SEARCH_ENDPOINT	Azure AI Search service endpoint url (e.g. https://YOUR_SEARCH_SERVICE.search.windows.net)
   SEARCH_API_KEY	Azure AI Search management key
   AOAI_KEY	Azure OpenAI service API key
   AOAI_ENDPOINT	Azure OpenAI service endpoint url (e.g. https://YOUR_AOAI.openai.azure.com/)

2. Rename .env.template to .env

3. Run ingest.py to add your PDF source to Azure Search index. This tool will perform parsing, chunking and batch indexation of the PDF document for you.

   e.g.
   python ingest.py -f PATH_TO_PDF_FILE

   

4. Run flow.dag.yaml to edit the RAG LLM App workflow where necessary.
   I recommend you to install the PromptFlow VS Code Extension for best local dev experience.

   

5. Deploy the LLM App to Azure ML Management Endpoint