This was the starting point of my public AI writing.
I wanted to show how to make a language model answer questions based on PDF content using Azure AI Search, Azure OpenAI, Blob Storage, embeddings, indexing, retrieval, and LangChain.
At the time, the tutorial was mostly a practical guide. Looking back, it also 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 eventually led me toward broader questions: when managed Azure services make sense, when open-source retrieval stacks are a better fit, and when fine-tuning is the wrong answer.
Read the original article:
Teach ChatGPT to Answer Questions: Using Azure AI Search & Azure OpenAI