“Generative AI is easy!” Look at ChatGPT, you ask a question, you get an answer! How difficult could it be?

In this 2-day hands-on workshop, we will cover the following topics in detail:

• An in-depth view of what is an LLM and what is its purpose is, including how it is built and made available for consumption.
• An in-depth view of Prompt Engineering as a tool that can help AI chatbots generate relevant and coherent responses in real-time conversations as It helps ensure the AI understands user queries and provides meaningful answers.
• Demonstration of Zero-Shot and Few-Shot Prompts
• Hands-on lab covering the use of Few-Shot prompts against OpenAI Large Language Models
• In-depth discussion on Orchestrators and the important role this provides and how to differentiate between the popular ones and when to use them:
  • Sematic Kernel - an open-source, AI Software Development Kit (SDK) from Microsoft. It allows developers to integrate large language models (LLMs) like Hugging Face, Azure OpenAI, and OpenAI with conventional programming languages like C#, Java, and Python. We will demonstrate how SK can help make working with LLMs more effective by marshaling and orchestrating inputs and functions. It also allows users to sanitize inputs, guiding the LLM to produce useful outputs.
  • LangChain - an open-source framework for developing applications using large language models (LLMs). LLMs are deep-learning models that are pre-trained on large amounts of data. They can generate responses to user queries, such as answering questions or creating images from text-based prompts. We will demonstrate its use in Orchestrating the communication and negotiation between your data and the LLM of choice through the Prompt.
• Hands-on lab covering the use of Semantic Kernal from Microsoft
• Hands-on lab covering the use of LangChain
• A deep dive into the Retrieval Augmentation process in an enterprise to give access to company data to be part of the prompt that will enrich the LLM to give more accurate responses.
  • RALM – Retrieval Augmentation Language Modeling.
  • RAG – Retrieval Augmented Generation.
  • ReAct - Reasoning and Action pattern.
  • COT – Chain of Thought pattern.
• Hands-on lab to exercise the concepts of RALM, RAG, ReAct and COT
• In depth view into Embedding Models and how they work as algorithms that convert high-dimensional data into low-dimensional vectors. This process is called dimensionality reduction, which simplifies data and makes it easier for machine learning algorithms to process.
• A view into the concept of Vectorization, also known as word embedding, and demonstrating the process of converting text data into numerical vectors. These vectors are used to build machine learning models where they can be used for Word Predictions, Similarities and Natural Languages.
• A look in the life of Vectorization Databases which is a type of database that stores data as mathematical representations, also known as vector embeddings. These mathematical representations are high-dimensional vectors, each with a certain number of dimensions. The number of dimensions can range from tens to thousands, depending on the complexity and granularity of the data. Discussion about some of the available ones like Azure AI Search, Pinecone, Chroma, Faiss, Azure Postgresql and others…
• Hands-on lab for using Embeddings and Vectorization databases.
• Explanation of what a Red Team and Security Hardening encompasses from a multi-layered attack simulation that uses a variety of tools, tactics, and strategies to breach defenses to reducing the attack surface for malicious actors.
• An overview of how Row level Security works in Generative AI as a security mechanism that limits access to specific rows in a database table. RLS is a feature in SQL that uses a user's role or permissions to restrict access to specific rows of a table. RLS is used to enforce security and privacy policies in a database and prevent unauthorized access to sensitive data and it can be used to limit access in a Generative AI environment using Database agents.
• Finally, we will discuss the importance of including Private Agents to the mix to allow different members of an organization to include private data only accessible to them in the prompt process of communicating with the LLM.

At the end of the workshop, attendees will have a working knowledge of how to work with LLMs and custom data using orchestrators while securing the environments and the data and provide scalability and extensibility.

Attendee Workstation Requirements:
You must provide your own computer (Windows or Mac) for this hands-on lab workshop with a camera, wired Internet connection, speakers, and a microphone.

Your computer must have the ability to run C# 10 code or Python: (One of Visual Studio 2022/VSCode 1.66+/VSMac/Rider/etc.) and your own Azure subscription and access to Azure OpenAI (request access to the Azure OpenAI preview and access to quota for GPT-3.5-Turbo and GPT-4 (optional) wand also Embeddings model for ada-002 for vectorization. The labs will be a mixture of Python and C#.