Teaching ChatGPT to Answer Questions with Azure AI Search

This was the starting point of my public AI writing.

The practical question was clear: how can a language model answer questions based on PDF content instead of only relying on its pretraining? The article walked through Azure AI Search, Azure OpenAI, Blob Storage, embeddings, indexing, retrieval, and LangChain to make that path concrete.

At the time, I thought of it mostly as a hands-on tutorial. Looking back, it introduced the theme I keep returning to: useful AI systems need grounding, retrieval, structure, and a workflow that developers can inspect.

That first RAG article led me toward broader architecture questions. When are managed Azure services the right fit? When does an open-source retrieval stack give more control? When is fine-tuning the wrong answer because the real problem is freshness, traceability, or retrieval quality?

The lesson I still take from this post is that document-grounded AI starts with a trust problem. The system has to show where its answers come from, not only produce fluent text.

Read the original article:

Teach ChatGPT to Answer Questions: Using Azure AI Search & Azure OpenAI