Course Overview
This 3-day Azure Open AI Hackathon is a platform for developers and AI enthusiasts to create innovative and intelligent AI-powered applications using Azure Open AI and other relevant Azure services. Participants will explore various scenarios, incorporating chat-based interfaces, natural language processing, image generation, and more to develop cutting-edge solutions powered by Open AI and Azure.
Course Objectives
This hackathon encompasses the modern approach to the world of generative AI – from a generic chatbot to a modern, AI-powered solution capable of extracting insights and generating content from various data sources. Each developer or AI enthusiast working with generative AI needs to have experience with these technologies to be successful in the exciting times in front of us. Every participant will build the following technical skills:
- Experience with Azure OpenAI platform (Azure OpenAI Studio, Azure AI Studio, creating and running custom chatbots, generating images using DALL-E, knowledge engineering)
- Experience with Azure AI Services (Azure Speech API, Azure AI Document Intelligence, Azure Vision API, Azure AI Search)
- Working knowledge with Azure platform in general (Azure SQL Database, Azure Blob Storage, Azure CosmosDB, Azure API Management)
Azure OpenAI Hackathon builds on the concept of learning by doing in a highly interactive environment. Solving real-world problems without being told directly what to do will enable participants to apply acquired skills immediately on their job.
Course Content
Challenges
The challenges are interconnected and build upon one another, allowing participants to develop an intelligent AI-powered application (alongside covering content filtering and embeddings), incorporate additional data by fine-tuning the models, and generate images with DALL-E using Azure Open AI and relevant Azure services. Participants will be part of a newly founded digital department in an up-and-rising retail corporation. They will be given access to Azure services and resources to address each challenge progressively.
Challenge 1: FAQs are so 20th century…
Participants will develop a modern documentation and support solution for a software product. Users can ask questions about the product and receive accurate answers and step-by-step explanations, leveraging Azure Open AI's Chat engine.
Azure AI Topics Covered:
- LLM Introduction
- Prompt engineering, construction of effective prompts
- System Prompt/Instructions
- Grounding
- Azure OpenAI Studio
- Azure Search (Basics)
Challenge 2: Onboarding made easier than ever
In this challenge, participants will create an internal knowledge base solution accessible through the chat-based web application. Leveraging Azure Cognitive Search, the application will efficiently retrieve relevant information in response to employees’ inquiries, making their onboarding a swift and pleasant experience.
Azure AI Topics covered
- Azure OpenAI SDK
- Azure AI Search (advanced)
- Azure AI Document Intelligence
- Grounding
Challenge 3: Paint me a picture
In this challenge, participants will enhance the AI-powered application by integrating Azure DALL-E for image generation. This will enable users to generate personalized avatars based on their inputs, such as country/city, age, favorite band, etc.
Azure AI Topics covered
- Azure OpenAI SDK
- DALL-E Image Generation
- Vision
Challenge 4: Maybe AI knows how to talk to our database?
Participants will create a basic chat-based interface using Azure Open AI's Completion engine. The focus is on integrating natural language understanding to allow users to interact with the database in a conversational manner. The chat-based application should understand and respond appropriately to user queries. Azure AI Topics covered
- Orchestration
- LangChain / PromptFlow / Semantic Kernel – options to choose from
Challenge 5: Which model do I ask?
Building upon the previous challenges, participants now have three different models: a general one (base model), one that interacts with teir data, and another that communicates with their database, in this challenge they will create a unified interface that allows users to interact with just one model by leveraging function calling and fine-tuning
Azure AI Topics covered
- Finetuning
- Combining different AI models for complex task execution
Challenge 6: The ultimate question – Safety!
Security in the LLM world is as vital as ever. In this challenge, participants will harden their Open AI solution by implementing better control mechanisms and protecting it against LLM-specific attacks (e.g., Jailbreaking, Prompt Injection, and Data Poisoning) leveraging Azure services.
Azure AI Topics covered
- Content Safety
- Content Filtering
- Jailbreak risk detection
Challenge 7: Enable seamless scalability of your services!
When building apps that leverage Azure OpenAI APIs, customers typically want more control over API access – authentication, throttling, monitoring, etc. The easiest way to achieve this is by using Azure API Management. In this challenge, participants will enhance their solution by implementing an access control layer to their instance of Azure OpenAI service, enabling only a certain number of requests per minute to be passed through to the service, depending on the user’s subscription level. In the second part of the challenge, they will replicate the solution to another region and implement failover mechanisms.
Azure AI Topics covered
- Azure API Management with Azure OpenAI
- Scalability, throttling, authentication, failover