Azure OpenAI LLM Cookbook

📌 Quick Reference: Curated Sample Collection

A one-stop hub, like a sample library. This repository is organized by topic to help reduce the time spent searching for and reviewing sample code. It offers a curated collection of minimal implementations and sample code from various sources.

[!IMPORTANT] 🔹For more details and the latest code updates, please refer to the original link provided in the README.app.md file within each directory.
🔹Disclaimer: Some examples are created for OpenAI-based APIs.

💡How to switch between OpenAI and Azure OpenAI endpoints with Python

• Programming Languages
  • Python:🐍
  • Jupyter Notebook:📔
  • JavaScript/TypeScript:🟦
  • Extra:🔴
• Status & Action
  • Created:✨ (A unique example found only in this repository)
  • Modified:🎡 (An example that has been modified from a referenced source)
  • Copied:🧲 (When created or modified emojis are not following)
  • See the details at the URL:🔗
• Microsoft libraries or products:🪟

⭐ If you find this repository useful, please consider giving it a star!

📖 Repository structure

📁 agent

• a2a_semantic_kernel🐍✨🔗🪟: Agent2Agent (A2A) Protocol Implementation with Semantic Kernel
• a2a_server_client🐍: Agent2Agent (A2A) Protocol - official implementation of Server/Client
• agent_multi-agent_pattern📔🪟: Agent multi-agent pattern
• agent_planning_pattern📔🪟: Agent planning pattern
• agent_react_pattern📔: Agent react pattern
• agent_reflection_pattern📔: Agent reflection pattern with LangGraph
• agent_reflection_pattern2📔: Agent reflection pattern (Groq as the LLM provider)
• agent_tool_use_pattern📔🪟: Agent tool use pattern
• arxiv_agent🐍✨🎡: ArXiv agent
• chess_agent🐍: Chess agent
• multi_agentic_system_simulator🐍✨🔗: A Multi-Agentic System Simulator. Visualize Agent interactions.
• role_playing📔: Role-playing
• web_scrap_agent🐍✨🎡🪟(Bing Search API): Web scraping agent
• x-ref: 📁industry

📁 azure

• azure_ai_foundry_sft_finetuning📔🪟: Supervised Fine-tuning
• azure_ai_foundry_workshop📔🪟: Azure AI Foundry Workshop
• azure_ai_search📔🪟: Chunking, Document Processing, Evaluation
• azure_bot📔🪟: Bot Service API
• azure_cosmos_db📔🪟: Cosmos DB as a Vector Database
• azure_cosmos_db_enn🐍✨🪟: Cosmos DB Exact Nearest Neighbor (ENN) Vector Search for Precise Retrieval
• azure_devops_(project_status_report)🐍✨🪟: Azure DevOps – Project Status Report
• azure_document_intelligence🐍🪟: Azure Document Intelligence
• azure_evaluation_sdk🐍🪟: Azure Evaluation SDK
• azure_machine_learning📔🪟: Azure Machine Learning
• azure_postgres_db📔🪟: pgvector for Vector Database
• azure_sql_db📔🪟: Azure SQL as a Vector Database
• copilot_studio🔗🪟: A low-code platform for bots and agents (formerly Power Virtual Agents)
• m365_agents_sdk🟦🪟: Rebranding of Azure Bot Framework
• sentinel_openai🔗🪟: Sentinel – Security Information and Event Management (SIEM)
• sharepoint_azure_function📔🪟: SharePoint Integration with Azure Functions
• teams_ai_sdk🔗🪟: Teams AI SDK

📁 cookbook

• anthropic: Anthropic Cookbook
• gemini: Gemini API Cookbook
• openai: OpenAI Cookbook

📁 data

• azure_oai_usage_stats_(power_bi)🔴🪟: Azure OpenAI usage stats using Power BI
• azure_ocr_scan_doc_to_table🐍✨🪟: Azure Document Intelligence – Extract tables from document images and convert them to Excel
• chain-of-thought🐍🔴: Chain-of-thought reasoning prompt
• fabric_cosmosdb_chat_analytics📔🔴✨(visual)🪟: Fabric: Data processing, ingestion, transformation, and reporting on a single platform
• firecrawl_(crawling)🐍: Firecrawl – Web crawling and scraping
• ms_graph_api📔🪟: Microsoft Graph API
• presidio_(redaction)📔🪟: Presidio – Data redaction and anonymization
• prompt_buddy_(power_app)🔴🪟: Prompt sharing application built on Power App
• prompt_leaked🔴: Prompt leakage detection and analysis
• sammo_(prompt_opt)📔🪟: A Structure-Aware Approach to Efficient Compile-Time Prompt Optimization
• semantic_chunking_(rag)📔: Semantic chunking for Retrieval-Augmented Generation (RAG)

📁 dev

• code_editor_(vscode)🐍✨🔗🪟: Visual Studio Code extension development
• diagram_to_infra_template_(bicep)🐍✨🪟: Bicep – Infrastructure as Code (IaC) language
• e2e_testing_agent📔🪟: End-to-end testing with Playwright automation framework
• git_repo_with_chat🐍✨: Chat with Github repository
• gui_automation🔗🪟: Omni Parser – Screen parsing tool / Windows Agent Arena (WAA)
• llm_router🐍✨🎡: LLM request routing and orchestration
• mcp_(model_context_protocol)🐍✨🔗: Model Context Protocol
• mcp_(sse)🐍✨🔗: Remote MCP (Model Context Protocol) calls
• mcp_to_openai_func_call🐍✨: MCP Tool Spec to OpenAI Function Call Converter
• memory_for_llm🐍🔗: Memory management techniques for LLMs – K-LaMP🪟
• memory_graphiti🐍✨: Graph and neo4j based Memory
• mini-copilot🐍✨🔗: DSL approach to calling the M365 API
• mixture_of_agents🐍✨🎡: Multi-agent system for collecting responses from multiple LLMs
• open_telemetry🐍✨: OpenTelemetry – Tracing LLM requests and logging

📁 eval

• evaluation📔: Using LLMs for automated evaluation and scoring
• guardrails📔: Guardrails for AI safety and compliance
• pyrit_(safety_eval)📔🪟: Python Risk Identification Tool

📁 framework

• agno_(framework)🐍: Agno – A simple, intuitive agent framework
• autogen_(framework)🐍🪟: AutoGen – A Framework for LLM Agent
• crewai_(framework)🐍: CrewAI – Agent collaboration framework
• dspy_(framework)🐍📔: DSPy – Declarative Language Model Calls into Self-Improving Pipelines
• guidance_(framework)📔🪟: Guidance – Prompt programming framework
• haystack_(framework)🐍📔: Haystack – NLP framework for RAG and search
• langchain_(framework)📔: LangChain – Framework for LLM applications
• llamaindex_(framework)📔: LlamaIndex – Data framework for LLM retrieval/agent
• magentic-one_(agent)🐍🪟: Magentic-One – Multi-agent system for solving open-ended web and file-based tasks
• mem0_(framework)🐍📔: Mem0 – LLM Memory
• omniparser_(gui)📔🪟: OmniParser – GUI automation and parsing tool
• prompt_flow_(framework)📔🪟: Prompt Flow – LLM Workflow
• prompty_(framework)🔗🔴🪟: Prompty – Prompt management
• pydantic_ai_(framework)🐍: Pydantic AI – Pydantic agent framework
• semantic_kernel_(framework)🐍🪟: Semantic Kernel – Microsoft LLM orchestration framework
• smolagent_(framework)🐍: SmolAgent – Hugging Face Lightweight AI agent framework
• tiny_troupe_(framework)📔🪟: Tiny Troupe – Multi agent persona simulation
• x-ref: 📁microsoft-frameworks-and-libraries:

📁 industry

• auto_insurance_claims📔: Automation for auto insurance claims processing
• career_assistant_agent📔: Career guidance and job recommendation agent
• contract_review📔: Legal contract analysis and review
• customer_support_agent📔: Customer support automation
• damage_insurance_claims📔: Automated claims processing for damage insurance
• invoice_sku_product_catalog_matching📔: Invoice and SKU reconciliation for accounting
• invoice_payments📔: Automation for invoice payments
• invoice_standardization📔: Standardizing invoice units for consistency
• music_compositor_agent📔: Music composition assistant
• news_summarization_agent📔: Automated summarization of news articles
• nyc_taxi_pickup_(ui)🐍: NYC taxi pickup analysis and UI visualization
• patient_case_summary📔: Summaries for patient medical cases
• project_management📔: a tools for project tracking and task management
• stock_analysis🐍✨🔗: AutoGen demo for analyzing stock investments
• travel_planning_agent📔: Travel itinerary planner
• youtube_summarize🐍✨: Summarizing YouTube videos using AI

📁 llm

• finetuning_grpo📔: Group Relative Policy Optimization (GRPO) for LLM fine-tuning
• knowledge_distillation📔: Compressing LLM knowledge into smaller models
• llama_finetuning_with_lora📔: LoRA – Low-Rank Adaptation of Large Language Models
• nanoGPT🐍: Lightweight GPT implementation
• nanoMoE🐍: Lightweight Mixture of Experts (MoE) implementation

📁 llmops

• azure_prompt_flow🔗🪟: Azure AI Foundry - Prompt flow: E2E development tools for creating LLM flows and evaluation
• mlflow📔: OSS platform managing ML workflows

📁 multimodal

• image_gen_dalle📔: Image creation with segmentaion
• openai-agents-sdk-voice-pipeline📔✨: OpenAI Agents SDK for voice processing
• openai-chat-vision📔: Multimodal chat with vision capabilities
• phi-series-cookbook_(slm)🔗🪟: Phi series models cookbook (small language models)
• video_understanding📔: Video content analysis and understanding
• vision_rag📔: Combining visual data with retrieval-augmented generation (RAG)
• visualize_embedding📔: Tools for embedding visualization and analysis
• voice_audio🟦: RTClient sample for using the Realtime API in voice applications

📁 nlp

• multilingual_translation_(co-op-translator)🐍🪟: a library for multilingual translation
• search_the_internet_and_summarize📔: Internet search and summarization
• sentiment_analysis_for_customer_feedback📔: Sentiment analysis for customer feedback
• translate_manga_into_english🐍✨: Manga translation into English
• txt2sql🐍: Converting natural language queries into SQL

📁 rag

• adaptive-rag📔: Adaptive retrieval-augmented generation (RAG)
• agentic_rag📔: Agent-based RAG system
• contextual_retrieval_(rag)📔: Context-aware retrieval for RAG
• corrective_rag📔: Improving retrieval results with corrective techniques
• fusion_retrieval_reranking_(rag)📔: Fusion-based retrieval and reranking for RAG
• graphrag📔🪟: Graph-based retrieval-augmented generation
• hyde_(rag)📔: Hypothetical Document Embeddings for better retrieval
• query_rewriting_(rag)📔: Enhancing RAG by rewriting queries for better retrieval
• raptor_(rag)📔: Recursive Abstractive Processing for Tree-Organized Retrieval
• self_rag📔: Self-improving retrieval-augmented generation

📁 research

• analysis_of_twitter_the-algorithm_source_code📔: Analyzing Twitter’s open-source ranking algorithm
• deep_research_langchain🐍📔: AI-driven deep research and analysis tools using LangChain
• deep_research_smolagents🐍📔: AI-driven deep research and analysis tools using smolagents
• openai_code_interpreter🐍📔: OpenAI’s code interpreter for data analysis
• r&d-agent🐍🪟: Research and development AI agent

🛠️ Comparing Local with Remote Repository

You can use the git_cmp.py script (and related files) to compare your local project directories with their corresponding remote GitHub repositories.

Typical Workflow

1. Index all projects and their GitHub URLs:

   python git_cmp.py --index --root <root_dir> --csv git_cmp_index.csv
   

   This creates a CSV file listing all projects and their remote URLs.

2. Compare local and remote repositories:

   python git_cmp.py --compare --root <root_dir> --csv git_cmp_index.csv --report git_cmp_report.txt --update_csv git_cmp_needs_update.csv
   

   This generates a report and a CSV of projects needing updates. It also copies changed files into .cache/ for review.

3. Update local files from cache (optional, use with care):

   python git_cmp.py --manipulate --root <root_dir> --update_csv git_cmp_needs_update.csv
   

   This copies files from .cache/ back into your project directories, optionally deleting files if flagged.

Mode & Command

• SUBDIR Mode
  target_remote_path refers to a path in a remote repository that should be used as the starting point for comparison.
  When the mode is set to SUBDIR, a directory with the same name as project_name is expected to exist under that path. The comparison should target all files and directories within that SUBDIR.
  If there are files or directories that exist only in the remote (excluding readme.app.md, .url, and any folder—including its subdirectories—that contains a DONOTCMP file), they should all be copied into the SUBDIR.

• FILE Mode
  When the mode is set to FILE, target_remote_path is not provided. A file with the same name is expected to exist at the specified GitHub URL. The comparison should target only that specific file, excluding readme.app.md, .url, and any folder—including its subdirectories—that contains a DONOTCMP file.

• ROOT Mode
  When the mode is set to ROOT, target_remote_path is also not provided. All files and directories under project_name (excluding readme.app.md, .url, and any folder—including its subdirectories—that contains a DONOTCMP file) should be compared with those in the specified GitHub URL. Files or directories that exist only in the remote should be copied into the project_name.

• Compare Command
  In compare mode, for files and directories that exist only in the remote, zero-size placeholder files will be created locally.
  Remote files and directories are treated as the source of truth.

• Manipulation Command
  In manipulation mode, the remote repository for project_name should be updated. It is first cloned using the git clone command into the .repo_cache directory. After cloning, the actual file copying process to the local directories begins.
  All local files and directories should be deleted first—excluding readme.app.md, .url, and any folder—including its subdirectories—that contains a DONOTCMP file—and then replaced with the corresponding remote versions. However, this local wipe will only be triggered when allow_delete == 'DELETE' to ensure safety.

• DONOTCMP is a marker file used to explicitly indicate that a directory is local-only and should be excluded from comparison.

Options

• --delay_sec <seconds>: Add a delay between GitHub API calls to avoid rate limits.
• --index: Index projects and write a CSV.
• --compare: Compare projects and write a report.
• --manipulate: Update local files from the cache based on the update CSV.

Example

python git_cmp.py --index --root . --csv git_cmp_index.csv
python git_cmp.py --compare --root . --csv git_cmp_index.csv --report git_cmp_report.txt --update_csv git_cmp_needs_update.csv
python git_cmp.py --manipulate --root . --update_csv git_cmp_needs_update.csv

Note:

• Create .env file. Set the GITHUB_TOKEN. e.g.,GITHUB_TOKEN=<your_key>
• Review .cache/ and the generated report before running --manipulate.
• See comments and docstrings in git_cmp.py for more details.

➡️ Convert ipynb to Python

• To convert a Jupyter notebook (.ipynb) into a runnable Python scrip

pip install nbformat nbconvert

import nbformat
from nbconvert import PythonExporter

# Load the notebook
notebook_filename = 'your_notebook.ipynb'
with open(notebook_filename, 'r', encoding='utf-8') as notebook_file:
notebook_content = nbformat.read(notebook_file, as_version=4)

# Convert the notebook to a Python script
python_exporter = PythonExporter()
python_code, _ = python_exporter.from_notebook_node(notebook_content)

# Save the converted Python code to a .py file
python_filename = notebook_filename.replace('.ipynb', '.py')
with open(python_filename, 'w', encoding='utf-8') as python_file:
python_file.write(python_code)

print(f"Notebook converted to Python script: {python_filename}")

💻 Microsoft Frameworks and Libraries

1. Semantic Kernel (Feb 2023): An open-source SDK for integrating AI services like OpenAI, Azure OpenAI, and Hugging Face with conventional programming languages such as C# and Python. It's an LLM orchestrator, similar to LangChain. / git
2. Azure ML Prompt Flow (Jun 2023): A visual designer for prompt crafting using Jinja as a prompt template language. / ref / git
3. SAMMO (Apr 2024): A general-purpose framework for prompt optimization. / ref
4. guidance (Nov 2022): A domain-specific language (DSL) for controlling large language models, focusing on model interaction and implementing the "Chain of Thought" technique.
5. Autogen (Mar 2023): A customizable and conversable agent framework. / ref / Autogen Studio (June 2024)
6. UFO (Mar 2024): A UI-focused agent for Windows OS interaction.
7. Prompty (Apr 2024): A template language for integrating prompts with LLMs and frameworks, enhancing prompt management and evaluation.
8. OmniParser (Sep 2024): A simple screen parsing tool towards pure vision based GUI agent.
9. TinyTroupe: LLM-powered multiagent persona simulation for imagination enhancement and business insights. [Mar 2024]
10. RD-Agent: open source R&D automation tool ref [Apr 2024]
11. Magentic-One: Built on AutoGen. A Generalist Multi-Agent System for Solving Complex Tasks [Nov 2024]
12. PyRIT (Dec 2023): Python Risk Identification Tool for generative AI, focusing on LLM robustness against issues like hallucination, bias, and harassment.
13. Presidio: Presidio (Origin from Latin praesidium ‘protection, garrison’). Context aware, pluggable and customizable data protection and de-identification SDK for text and images. [Oct 2019]
14. Microsoft Fabric: Fabric integrates technologies like Azure Data Factory, Azure Synapse Analytics, and Power BI into a single unified product [May 2023]

📚 References & Sources

• OpenAI Cookbook
• LangChain Cookbook
• LlamaCloud Demo
• Chainlit Cookbook
• Microsoft AI Agents for Beginners
• GenAI Agents by NirDiamant
• RAG Techniques by NirDiamant
• Gemini API Cookbook
• Anthropic Cookbook
• Awesome LLM Apps
• AI Engineering Hub

Contributor 👀

ⓒ https://github.com/kimtth all rights reserved.