Idea to app
1. Idea to app
SPEAKER: Given that foundation models are the backbone of generative AI development and applications, you may wonder how to interact with them and bring your ideas to applications. Are there tools available to assist with this process? Let's look at a use case that may resonate with your real-world problems. Bea, Ann, and Ian, all work for Cymbal Insurance, a national insurance company with a strong presence in the Western states. They are looking for Gen AI tools to help them in their everyday work, including conducting research and automating their workflow. Bea, business analyst-- seeks to quickly prototype a Gen AI app idea that automates risk analysis and report generation, despite lacking a technical background. Ann, AI developer-- needs a user-friendly development platform for prompt engineering, including drafting, evaluating, refining, and managing prompts. Ian, ML engineer-- requires a robust, secure, and scalable tool to build pipelines for deploying prompts to production and fine-tuning Gen AI models. Powered by advanced foundation models like Gemini and an enterprise-ready AI infrastructure, Google offers a variety of products and services that can help Bea, Ann, and Ian accomplish their Gen AI use cases. For example, Vertex AI Studio-- easily build, deploy, and scale generative AI applications. Agent Builder and Gemini Enterprise-- design, deploy, and manage AI agents. NotebookLM-- an AI-powered research and note-taking tool for document interaction and insights. Let's check out Vertex AI Studio first and then move to AI Agents and NotebookLM later in this course. What is Vertex AI Studio? Simply put, Vertex AI Studio is your gateway to generative AI. Vertex AI Studio provides an intuitive interface between developers and the foundation models. It enables you to build Gen AI applications in a low-code, or even no-code environment where you can rapidly test and prototype applications, tune and customize models using your own data, augment them with real-world, up-to-date information, and deploy models efficiently in production environments with auto-generated code. Envision Vertex AI Studio as a cutting-edge workshop where Gen AI models are your raw materials. You are the craftsperson, and the Vertex AI Studio toolkit is your arsenal for shaping and refining these models into powerful AI solutions. Intrigued? Join Bea, Ann, and Ian to learn how to use this magical tool from prompt to production. They all understand that the prompt-to-production may involve a comprehensive lifecycle. Designing, evaluating, and refining prompts, building and testing applications, and monitoring and optimizing generative AI models. However, Bea, with no technical background, wonders if there's a quicker way to directly turn idea to app. The journey begins with a prompt, a natural language request to an AI model. This can be a question, task, or instruction leading the AI to generate text, code, images, videos, music, or more. The process of creating prompts to get the desired response is called prompt design. The iterative process of designing, refining, and optimizing prompts to effectively guide an AI model in generating desired and high quality outputs is called prompt engineering. Think of a prompt as the way to communicate with Gen AI models. Just as human communication requires clarity, so does prompting AI. You need to be good at asking questions to get the results you want. So what makes a good prompt? Let's begin by examining the anatomy of a prompt. Generally, a prompt includes one or more of the following key components-- task, context, examples. Task is required. This is the core instruction for the model. For example, "Conduct a risk analysis for an insurance company." Simple tasks may only require zero-shot prompting, which means providing only the task without any examples. Context is optional. This is the background information or system instructions that sets the stage for the AI, such as, "You are a business analyst overseeing risk assessment for an insurance company." Examples are optional. These are demonstrations of desired responses, step-by-step instructions or output formats that are useful for complex tasks, such as guiding the AI with a report template. This is also known as few-shot prompting. When crafting effective prompts, focus on two key aspects, content and structure. For content, ensure your prompt includes all relevant information for the task, such as clear instructions, context, and examples. For structure, organize the information in a way that the model can understand. Consider the order, labels, and delimiters. Here's an example of a well-structured prompt. You first describe the context. "You are an IT help desk technician at a university. Your daily job is to help faculty and students solve their technology issues." You then specify the task by providing a step-by-step instruction, such as, "To complete the task, you will need to follow these steps." Additionally, you also provide some common Q&A examples. Tips for effective prompts-- now that you understand the ingredients of a good prompt, here are some tips for crafting effective ones. Be direct and specific. State requests clearly, and use keywords. Use structure. Break down complex tasks into smaller steps, and use delimiters to organize sections. Iterate and refine. Start simple, and improve based on AI output. Explore advanced techniques. Consider few-shot prompting, chain-of-thought prompting, or Retrieval Augmented Generation, or RAG, for more complex scenarios. Some of these advanced techniques will be discussed later in this course. And remember basics-- avoid jargon, set clear goals, create scenarios, and encourage analysis. As Bea and Ann reflect on their discovery journey with Vertex AI Studio, they want to create a prompt that utilizes key components and best practices. Which of the following prompts is the best option? A, provide a risk assessment report. B, conduct a market risk analysis for a health insurance company in the United States. C, you are an analyst at a regional health insurance provider in the southeastern United States. Your task is to generate a market risk analysis by following the steps A, B, and C. Please find the report template that includes 1, 2, and 3. Yes, C is the correct answer. Take a moment to think about why. What makes C an effective prompt? Compared to A and B, C clearly outlines all three components-- task, generate a market risk analysis; context, you are an analyst at an insurance company; and examples, such as steps and template. This detailed instruction will effectively guide the AI. Excited, Bea and Ann then used Vertex AI Studio to prototype a web-based application. Bea is a little anxious about her first prompt. But Vertex AI studios Help me write feature provides AI-assisted prompting, clarifying content, formatting responses, and breaking down complex tasks. The platform's prompt gallery also offers numerous examples, filtered by modality such as audio, doc, text, image, and video, tasks, such as answer questions, classify, and code, and features. Ann is particularly impressed by Vertex AI studio's support for multimodal prompts and outputs, allowing embedding documents, PDFs, images, videos, and YouTube content in prompts and generating responses in similar multimodal formats. With the AI assistant and prompt gallery's help, Bea drafted her first confident prompt. "Conduct a risk assessment on housing in southern Los Angeles. You are a business analyst for Cymbal Insurance. Analyze the articles from the internet, and extract the following information. Risk assessment-- identify potential risks and rate severity 1 to 5, low to high. Categorization-- classify risks by geography, type, and sentiment. Impact analysis-- evaluate potential consequences of each risk. And additional insights. Provide relevant observations and recommendations." Take a moment to reflect on Bea's prompt. Are you able to identify the major components that make up an effective prompt-- task, context and examples? Do you have any suggestions to improve it? With a few rounds of experimenting with prompts, Bea and Ann are ready to see their first prototype. They click on the Build with Code button and Deploy as App. And voila! Vertex AI Studio automatically generates a web-based application. Bea and Ann are amazed at how quickly they were able to prototype an idea and discover the capabilities of Gen AI. They can't wait to see more options provided by Vertex AI Studio and dive deeper to design, evaluate, and refine prompts. You'll learn more about this soon.2. Let's practice!
Create Your Free Account
or
By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.