Azure AI Apps

A collection of Python demo applications for engineers learning how to build Retrieval-Augmented Generation (RAG) pipelines, AI agent orchestration, and chatbot UIs on Azure AI services. One key focus is to show how to leverage AI hosting capabilities on Azure Kubernetes such as KAITO inference AI models and hosting the applications in AKS.

These are demo and learning projects for myself that I blog about, use as demos for online and in-person tech presentations and show case in my youtube channel https://youtube.com/roykimyyz

Each sub-project is a standalone, runnable demo that highlights a different architecture pattern or Azure service integration — from simple RAG chat to multi-agent frameworks deployed on Azure Kubernetes.

---

Table of Contents

#	Project	What You'll Learn	Key Technologies
1	agent-rag-resume	Building a RAG agent with Semantic Kernel plugin system	Semantic Kernel, Azure AI Search, Streamlit, Azure Kubernetes Service
2	agentframework	Streamlit chatbot with Multi-agent orchestration with Microsoft Agent Framework SDK on AKS	Agent Framework SDK, KAITO, AKS
3	app-rag-aisearch-sk	SK vector store connectors, memory plugins, and search skillset pipelines	Semantic Kernel (plugins + memory), Azure AI Search, Flask
4	rag-app-resumes	End-to-end RAG indexer pipeline with skillsets and Prompty templates	Azure AI Projects SDK, Prompty, OpenTelemetry
5	rag-chatapp-retail	RAG with AI evaluation (groundedness metrics)	Azure AI Evaluation SDK, Prompty
6	streamlit-chat	Streamlit chat UI with basic and agentic variants. Basic implementation.	Streamlit, Semantic Kernel, Azure AI Agents
7	python-uv-project-template	Starter template for uv-based Python projects	uv, Ruff, mypy

---

Core Azure Services Used

Service	Role in These Demos
Azure OpenAI	Chat completions (GPT-4o / GPT-4o-mini) and embeddings (text-embedding-ada-002 / text-embedding-3-large)
Azure AI Search	Document indexing, vector/hybrid/semantic search with skillsets and indexers
Azure AI Foundry	Project management, connection handling, and AI inference via AIProjectClient
Azure Blob Storage	Source document storage (resume PDFs, product data)
Azure Kubernetes Service (AKS)	Container hosting with KAITO GPU inference and Ingress
Azure Container Registry (ACR)	Docker image storage and deployment
Azure Monitor / Application Insights	OpenTelemetry tracing and telemetry

---

Sub-Projects

1. agent-rag-resume

📁 agent-rag-resume/

What it demonstrates: A RAG chatbot that searches indexed resume documents using Azure AI Search and generates grounded, cited answers via Semantic Kernel's plugin system.

How it works: Embeds the user's question → performs vector search against a pre-built resume index → passes retrieved document chunks as grounding context to the LLM → returns cited answers with [title] references.

Technologies	Semantic Kernel (≥ 1.0.3), Azure OpenAI, Azure AI Search, Streamlit, azure-ai-projects
Package Manager	uv
Entry Points	agent_rag_resume/ai_foundry_agent_sk.py — SK version agent_rag_resume/azure_rag_chat_ai_foundry_agent_style_python_single_file_app.py — direct SDK version
Key Concepts	SK prompt plugins (UserPromptPlugin), grounded answers with citations, two implementation variants for comparison
Deployment	AKS manifests included (Kustomize)

cd agent-rag-resume
uv sync && uv run streamlit run agent_rag_resume/ai_foundry_agent_sk.py

---

2. agentframework

📁 agentframework/

What it demonstrates: Multi-agent AI chatbot built with Microsoft's Agent Framework SDK, showing how to orchestrate multiple LLM backends through a single config-driven architecture.

How it works: Uses the SDK's ChatAgent orchestration loop and BaseChatClient contract to route conversations to different backends (Azure AI Foundry, KAITO GPU inference, KAITO RAGEngine) based on YAML configuration.

Technologies	Agent Framework SDK (v1.0.0b), Streamlit, Jinja2, PyYAML, Pydantic, Click
Package Manager	uv · Python 3.13
Entry Points	chatbot/chatbot.py — Streamlit UI cli.py — CLI agents ai_chat_client.py — BaseChatClient implementation
Key Concepts	5 agents across 3 backends, Strategy pattern for backend swapping, custom KaitoChatClient, config-driven agent selection
Deployment	Production K8s (Kustomize, Ingress-NGINX, HPA)

cd agentframework
uv sync && uv run streamlit run chatbot/chatbot.py

---

3. app-rag-aisearch-sk

📁 app-rag-aisearch-sk/

What it demonstrates: RAG application powered by Semantic Kernel's full plugin ecosystem — vector store connectors, ChatCompletionAgent, TextMemoryPlugin, and Prompty prompt templates for document and resume chat.

How it works: Uses SK's AzureAISearchCollection vector store connector to query indexed documents, maintains conversation history via memory plugins, and generates answers through SK's ChatCompletionAgent.

Technologies	Semantic Kernel (with Azure extensions + memory), Azure AI Search, Azure OpenAI, Flask, Prompty
Package Manager	pip (requirements.txt)
Entry Points	chat_with_docs.py — SK document chat chat_with_resumes.py — resume RAG chat chat_with_docs_api.py / chat_with_resumes_api.py — Flask APIs create_skillset_run_indexerv2.py — indexer pipeline setup
Key Concepts	SK vector store connectors, SK prompt plugins, memory plugins, index/skillset/indexer pipeline creation, Prompty templates

cd app-rag-aisearch-sk
pip install -r requirements.txt
python chat_with_docs.py
# or Flask API:
python chat_with_resumes_api.py  # http://localhost:5000

---

4. rag-app-resumes

📁 rag-app-resumes/

What it demonstrates: End-to-end RAG pipeline for tech recruiters — from PDF upload to Azure Blob Storage, through indexing with Azure AI Search skillsets, to natural-language Q&A with grounded, cited answers.

How it works: Upload resume PDFs → index with Azure AI Search (SplitSkill → EmbeddingSkill → EntityRecognitionSkill) → hybrid vector + keyword search → generate grounded answers using Azure AI Projects SDK + Prompty templates.

Technologies	Azure AI Projects SDK, Azure AI Inference SDK, azure-search-documents, Prompty, Flask, OpenTelemetry, pandas
Package Manager	pip (requirements.txt)
Entry Points	chat_with_resumes.py — CLI chat_with_resumes_api.py — Flask API (/api/chat) create_skillset_run_indexerv2.py — indexer pipeline upload-data-blobstorage.py — PDF upload
Key Concepts	Full indexer pipeline (SplitSkill → EmbeddingSkill → EntityRecognitionSkill), Prompty templates for intent mapping and grounded chat, hybrid search, OpenTelemetry tracing

cd rag-app-resumes
pip install -r requirements.txt
python chat_with_resumes.py --query "Find candidates with Python and Azure experience"

---

5. rag-chatapp-retail

📁 rag-chatapp-retail/

What it demonstrates: RAG for retail product search with an AI evaluation pipeline — measures answer groundedness against source documents using Azure AI Evaluation SDK.

How it works: Maps user intent → generates embeddings → performs vector search against a product index → generates grounded answers → evaluates response quality with GroundednessEvaluator.

Technologies	Azure AI Projects SDK, Azure AI Evaluation SDK, Prompty, Flask, OpenTelemetry, pandas
Package Manager	pip (requirements.txt)
Entry Points	chat_with_products.py — product RAG chat get_product_documents.py — intent → embed → vector search evaluate.py — groundedness evaluation create_search_index.py — index creation
Key Concepts	Intent mapping with Prompty, VectorizedQuery product search, GroundednessEvaluator evaluation pipeline, chat protocol compliant responses

cd rag-chatapp-retail
pip install -r requirements.txt
python chat_with_products.py --query "I need a new tent for 4 people"
python evaluate.py  # Run groundedness evaluation

---

6. streamlit-chat

📁 streamlit-chat/

What it demonstrates: Streamlit-based ChatGPT-like web UI for querying indexed documents, with both a basic and an agentic variant using Semantic Kernel memory and Azure AI Agents.

How it works: Provides a chat interface that connects to Azure OpenAI and Azure AI Search. The agentic version adds SK TextMemoryPlugin with AzureCognitiveSearchMemoryStore and Azure AI Agent integration.

Technologies	Streamlit, OpenAI Python SDK, Semantic Kernel (with Azure memory), azure-search-documents, Prompty
Package Manager	pip (requirements.txt)
Entry Points	chatbot.py — basic chatbot UI chatbot-agentic.py — SK + Azure AI Agent version agent.py — standalone agent script
Key Concepts	Chat UI patterns, sidebar settings (model/endpoint/search toggle), SK memory store integration, Azure AI Agent integration

cd streamlit-chat
pip install -r requirements.txt
streamlit run chatbot.py --server.port 8502

---

7. python-uv-project-template

📁 python-uv-project-template/

What it demonstrates: Minimal starter template for new Python projects using the uv package manager with Ruff linting and mypy type checking.

Technologies	Python 3.13, uv, Ruff, mypy
Entry Points	main.py
Key Concepts	pyproject.toml configuration, uv workflow, Ruff + mypy setup

cd python-uv-project-template
uv sync && uv run main.py

---

Getting Started

Prerequisites

• Python 3.12+ (3.13 for agentframework)
• uv package manager (for uv-based projects)
• Azure CLI (az) with an active subscription
• An Azure OpenAI resource with chat and embedding model deployments
• An Azure AI Search resource

Setup Pattern

Each sub-project follows a similar workflow:

1. Navigate to the sub-project directory
2. Install dependencies — uv sync (uv projects) or pip install -r requirements.txt (pip projects)
3. Configure environment — copy .env-sample to .env and fill in your Azure resource credentials
4. Run — see the commands and entry points listed above

💡 Tip: Each sub-project has its own README with more detailed setup instructions.

---

License

This project is for educational and demonstration purposes.