RAG solution without Azure AI Search

[gkarwchan]

All RAG tutorial I saw on Azure Foundry are using Azure Search AI. But that service is expensive.
For developers who want to learn the RAG, is there a cheaper alternative?

[PhxCustomerInsights]

The only way is to create a new Azure account, and you get 200 in credentials.

Look at the new pricing for ContentUnderstanding, but the Azure AI Search is way to expense - last month for my learning I paid over $300 per month for a few simply queries to my own RAG application.

[AbeOmor]

Hi @gkarwchan, I'm a PM work on the Azure Postgres focused on integrations with Azure AI foundry for RAG and Agentic AI.

Azure PostgreSQL combined with the pgvector extension can serve as a cheaper alternative to Azure AI Search for building Retrieval-Augmented Generation (RAG) systems.

1. Microsoft Learn Path on using RAG and AI foundry with Postgres:
2. Video Series for Build AI apps with Postgres
3. Docs for GenAI in Postgres

What is pgvector?

• pgvector is an open-source extension for PostgreSQL that adds support for storing and searching vector embeddings.
• You can store embeddings (from models like OpenAI, Hugging Face, etc.) directly in a PostgreSQL table using a vector data type.
• It allows similarity searches via:
  • Cosine similarity.
  • Inner product.
  • Euclidean distance.

Why It’s Cheaper

• PostgreSQL is free (open-source) and can run on any local machine, self-hosted server, or on a cloud service like Azure Postgres Flexible Server.
• You’re only paying for basic compute and storage. Also you pay for the cost of the models you use from Azure AI Foundry
• No query costs per embedding search.

[gkarwchan]

Thanks a ton