With so much news around AI RAG – what’s the easiest way to get started to “play around” with AI and PDFs? While there isn’t one definitive answer I’m going to walk through one using Azure AI Studio. This example is good when:
- You have an existing Azure subscription and aren’t watching the costs too closely
- You want something you can easily click around and try some options without having to set up lots of tooling, pipelines, code etc.
Prerequisites
- Microsoft Azure Subscription – free trial here
- Microsoft Azure Open AI resource – how to create a resource here
- Documents/text you want to use
Create your Azure AI Studio project
Go to https://ai.azure.com/ and create yourself a new project
Deploy an embedding – There’s a few to choose from but I picked the ada-002
Deploy a chat completion – There’s a few to choose from. If you want to use them in the chat playground I tend to pick gpt 35 turbo or gpt 4o mini
Generally I’ll pick the defaults, but check your token limit. Some default to 0 which will mean you wont’ get very far using it. It’s easy to update afterwards as you need
Create an index using documents on your computer
Set your search service or create a new one – I set up the standard tier. For ease of finding the right index later give your index a meaningful name or if you like to live dangerously leave the generated name
We want to use vector search so ensure you choose that option and pick the embeddings we added earlier or let it create one if you missed the step above
Depending on how big your documents are this could take awhile.
When that’s finished, got the chat playground, choose your desired chat deployment, the index you made and how you’d like to search and off you go.
What I like about the playground is it’s quick to change settings, ask some basic questions and see the references:
2 Responses