Over the last 11 months, I have been steadily been building various AI apps to leverage features and capabilities offered by Azure services that support AI workloads. This includes Azure Open AI service, Azure AI Foundry, Azure Kubernetes, Kubernetes AI Toolchain Operator (KAITO), Azure AI Search and more.

Visit my github repo at https://github.com/RoyKimYYZ/azure-ai-apps.

This repository (azureai-chatapp) documents a hands-on, progressive learning journey through generative AI application patterns on Azure and some leveraging Azure Kubernetes. Each subfolder is an independent project that explores a specific pattern and technology stack. I am also building on skills and patterns from the previous iteration.

My journey began with simple Azure Open AI text completions with python. But this repo started with its first project in March 2025.

• rag-chatapp-retail – A product-catalog chatbot where users ask questions such as “I need a tent for 4 people”. Product data from a CSV file is vectorized with Azure OpenAI embeddings (text-embedding-ada-002), then indexed into Azure AI Search, and the GPT-4o model generates grounded answers from retrieved search results (i.e. text chunks). This project also introduced Prompty templates for prompt engineering. This project as modeled after a MS Learn tutorial.
• rag-app-resumes – An early prototype for resume search that was later superseded. Then use case is for tech recruiters to upload and search for resumes. Deployed onto AKS.
• streamlit-chat — A Streamlit web UI providing a chatbot-style interface that calls the RAG APIs. This was the first front-end layer. This was my first exploration into using Streamlit as a chatbot. I thought it was really cool and gets you off the ground running. I loved the side panels for selecting configuration settings on the fly.

Key skills learned: Azure AI Search (index creation, indexers, skillsets, datasources), Azure OpenAI embeddings + chat completions, vector + semantic hybrid search, Prompty prompt templates, Flask REST APIs and RAG evaluation pipelines.

Semantic Kernel & Advanced RAG

•  app-rag-aisearch-sk – By August 2025, the focus shifted to Microsoft Semantic Kernel. I wanted to learn an AI orchestration framework that introduces plugins, prompt templates, and agent abstractions against AI Foundry and Azure Open AI endpoints. This project applied Semantic Kernel to the resume-search domain for tech recruiters.
  Built a full Azure AI Search pipeline: blob storage ingestion of PDF resumes -> chunking skillsets -> indexer -> vector index with semantic ranking.

Key skills learned: Semantic Kernel SDK (Kernel, agents, plugins, function calling), Azure Blob Storage integration, Azure Kubernetes Hosting, AI Search indexer/skillset pipelines, SK prompt template patterns, structured data models for resume parsing.

AI Foundry Agents & Prompt Engineering

• agent-rag-resume – In September 2025, my experimentation moved to Azure AI Foundry Agents exploring two approaches side by side:
  A “traditional” single-file Python agent using Azure AI Foundry’s agent API directly.
  A Semantic Kernel–based agent (ai_foundry_agent_sk.py) with custom UserPromptPlugin prompt templates to control agent behavior via Semantic Kernel’s plugin system.

Multi-Agent Framework on AKS with KAITO (Jan–Feb 2026)

• agentframework – January 2026I want to explore the successor to Semantic Kernel by building a Multi-agent chatbot with the new Microsoft Agent Framework.
• Three LLM backends through app platform:
  • Azure AI Foundry – GPT-5.2 and DeepSeek-V3.2
  • KAITO inference model – phi-4-mini-instruct running on in-cluster GPU nodes on AKS
  • KAITO RAGEngine – AKS hosted RAG on a GPU nodepool.

If you are wondering what tech engineers have been doing to learn and apply generative AI applications on Azure, I hope my story has given some insight. Hope this help you give expousre and direction on how to map out your learning journey on these exciting technologies. Please check out my repo: https://github.com/RoyKimYYZ/azure-ai-apps.