Generative AI for Java Developers with Azure OpenAI ChatGPT

Explore the concept of intelligence and show how artificial intelligence simulates human thinking using data from our senses and learning from our experiences to solve new problems with models.

Explore how generative AI works, how it differs from traditional AI and machine learning, and the roles of NLP, GPT models, and transformer architectures in generating new content.

Explore how OpenAI builds models to tackle real world problems in text, images, and audio. Learn why the OpenAI API enables integration via RESTful HTTP with Python and JavaScript libraries.

Discover how ChatGPT uses OpenAI's large language models like GPT to generate intelligent responses from prompts for tasks from questions to campaigns, and note its limitations.

Set up and explore ChatGPT essentials, including choosing GPT-3.5 or GPT-4, using the chat interface, saving and sharing chats, and configuring custom instructions, data controls, and prompts.

Learn prompt engineering as guiding a model with carefully designed prompts. Use action verbs and four guidelines—context, audience, style/format, and output length—to tailor clear, precise outputs.

Learn to use ChatGPT for text summarization with prompt engineering, output limits, and focus. See a practical Ducati Diavel example showing how to structure prompts for concise, meaningful results.

Explore text expansion, generating content from ideas and shaping tone, length, structure, and audience. See prompts that turn brief job descriptions and informal emails into professional content.

Learn to research and plan with ChatGPT by using a travel planner persona and multi-step prompts to craft an eight-day Europe itinerary from Stuttgart to London, with budgets and updates.

Explore how ChatGPT acts as a universal programmer across languages, generating, documenting, explaining complex code, and fixing issues, with Python examples that filter IT courses by rating above four.

Learn how to use ChatGPT to classify customer service tickets into technical issues, billing queries, and product feedback, with prompt inputs and role settings for accurate routing.

See how ChatGPT enables marketing by generating a complete film campaign—from teaser and trailers to social media and influencer partnerships—and craft a targeted marketing email for a generative AI course.

Learn how ChatGPT serves as a global translator, understanding any language and translating to any language with prompts and the translation window, from Spanish to English or Hindi.

Apply conditional prompting to generate a title only for English text, handle Spanish inputs with translations, and use multi-condition prompts to control when a title appears.

Learn to command output structure in ChatGPT, producing numbered or unordered lists, tables with custom columns, and structured paragraphs with headings for clear, reusable AI responses.

Learn to generate sample data with ChatGPT, creating a ten-record CSV (id, name, age, email, phone) and converting it to a JSON array for RESTful APIs.

Explore how the Azure OpenAI service provides access to OpenAI models such as GPT, Codex, and Dall-E embedding models, with enterprise security, private networking, RESTful APIs, and Azure OpenAI Studio.

Submit an access request for the Azure OpenAI service, select text, code, Dall-E, OpenAI, and whisper models, and ensure you have a corporate Azure subscription and email.

Navigate the Azure OpenAI service documentation and model list, and understand regional availability for GPT-4, GPT-3.5, embeddings, and fine tuning.

Access deployed large language models via Azure OpenAI RESTful endpoints by adding a ready-to-use library as a dependency, enabling the OpenAI client to access text, image, audio, and embedding models.

Explore pricing, quotas, and limits for Azure OpenAI services, including token costs and context. Compare GPT-3.5 turbo and GPT-4 pricing, fine tuning, hosting, DALL-E images, whisper, and regional quotas.

Deploy an Azure OpenAI resource in the portal, choose East US for text and image models, set a unique name with standard pricing, then open the Azure OpenAI Studio.

Explore Azure OpenAI Studio to deploy, test, and customize OpenAI models, including GPT-3.5-turbo, text embeddings, DALL-E, and audio models, then train with your data.

Deploy a text model in the Azure OpenAI Studio by creating a new deployment of GPT-3.5 turbo named text_demo. Configure quotas, content filters, and open the deployment in the playground.

Learn to fine-tune model behavior in the chat playground by configuring a system message, examples, and prompts, manage chat sessions, panels, deployment, and history.

Learn to control model behavior by adjusting max response, temperature, top P, stop sequences, and penalties, and manage tokens within a 4000-token limit across system messages, examples, prompts, and responses.

Install postman and launch it to test your microservices RESTful application programming interface. Use http methods like get, post, put, and patch, and configure headers, json bodies, and basic authentication.

Access the deployed OpenAI model from postman by posting to a restful endpoint with an api key header and a json body using GPT-3.5-turbo.

Set up a Maven Java project named 'Azure Open AI demo' with a Maven Archetype, configure Java 21, add the Azure Open AI Maven dependency, and reload the project.

Set up an environment variable named openai_api_key for the OpenAI key on Windows or Mac, then access it in Python with os.getenv and export it in your bash profile.

Walk through creating an OpenAI client with the Azure endpoint and API key, building chat messages, and configuring chat completion options before calling get chat completions on Azure-deployed models.

Learn how to call the Azure OpenAI chat completion API from a Java project by configuring the endpoint, credentials, and system and user messages to retrieve the assistant response.

Learn to fetch multiple responses for a single prompt using chat completions options, iterating over choices and printing each message with its role and content.

Deploy a trained model as a web app in Azure with Azure Active Directory authentication for a secure chatbot, and configure basic pricing and a Linux app service.

Explore the completions playground in Azure OpenAI, using the old completion API to generate emails with subject and body from a single prompt, and compare it to chat-based workflows.

Azure OpenAI content filtering detects harmful prompts and outputs across hate, sexual, self-harm, and violence, with safe to high severities, a default medium level, and customizable filters.

Create customized content filters in Azure OpenAI to block prompts and model outputs across hate, sexual content, self-harm, and violence. Apply the filter to deployments and test with completions.

Generate and edit images with Dall-E by providing plain English prompts. Control variations and size, and use original plus masked images for edits via the create_edit method.

Learn to generate images with DALL-E in Azure OpenAI Studio, use the DALL-E playground and prompts, and run Python code to create and size up to three images.

Generate Zurich cityscape images with the DALL-E API in a Java project by setting image generation options and retrieving image URLs showing reflections in the lake via Azure AI.

Explore additional image generation options, set the number of images and size, and learn to handle errors when DALL-E returns a URL or an error.

Explore how text embeddings convert words and sentences into numeric vectors that reveal meaning and relationships, enabling document similarity, search, recommendations, and translation using OpenAI embeddings and cosine similarity.

Deploy embeddings model in Azure OpenAI Studio by creating a deployment, selecting the text-embedding option, and naming it embed_underscore_demo to obtain embeddings for text, words, and sentences via the API.

Calculate word and sentence embeddings in a Java program by calling the Azure deployed embedding model, iterating results, and printing the embedding vectors.

Learn to compute semantic similarity between words and sentences using embeddings and cosine similarity, by comparing embeddings for good and bad and exploring Azure semantic search.

Explore Whisper, an OpenAI audio model that transcribes and translates speech to text, supports mp3 to webm and a 25 mb limit, and enables Python-based transcription via the Transcribe method.

Deploy a new azure openai resource in the west europe region to access the whisper model. Open the azure openai studio to verify whisper alongside gpt and text embedding models.

Deploy a text model and a whisper audio model in Azure OpenAI Studio, naming them text_underscore_demo and audio_underscore_demo. Update the new keys in your environment variables to avoid access denied.

Learn to transcribe and translate audio with Whisper Playground in Azure OpenAI, upload MP3s, choose transcribe or translate, and auto-detect language for quick text.

Update your audio resources by unzipping audio.zip to access three audio files, then copy the audio directory to a convenient folder and note the path for upcoming lectures.

Transcribe audio to text using whisper models in a Java app with Azure OpenAI APIs. Configure paths and keys to generate json subtitles from an AWS Lambda audio.

Translate a French audio to English and print the translation with transcription using the whisper model. Rename the transcriber to translator and switch to audio translation options in the code.

Learn to mix audio and text models by transcribing audio with whisper, then summarize the transcription into two bullet points using a single OpenAI client.

Learn how to use your data with Azure OpenAI, ingest data into storage and cognitive search, and generate model responses with prompt context for chat bots and insights.

Learn how Azure OpenAI ingests data, stores it in Azure storage, and builds a cognitive search index to retrieve relevant text for prompts, using keyword, vector, or hybrid search.

Learn how ingestion works in Azure OpenAI: upload data via Azure Blob storage, Cognitive Search index, or direct upload; chunk into 1024-token pieces, then index and embed for vector search.

Load five resumes into the chat bot via add your data, then test queries in the playground to help HR filter data scientist candidates with Python skills.

Create a storage account with a data container in the same resource group as the OpenAI service in East US, then deploy and delete cognitive search service to avoid charges.

In this lecture, learn to ingest your data into an Azure cognitive search keyword search pipeline: enable cors, create an index, upload files, and query with adjustable strictness.

Explore how hybrid search blends keyword and vector search by deploying an embedding model, creating a neyah vector hyphen index, and running hybrid queries over uploaded data.

Develop a smart home hub chatbot using generative ai and Azure OpenAI ChatGPT, feeding data to a text model with cognitive and hybrid search to answer compatibility and integration questions.

Learn how to clean up cloud resources by deleting the cognitive search resource and storage container, confirming deletions to manage costs in Azure OpenAI workflows.

Explore fine tuning, retraining a base model with task-specific data to improve performance and incorporate new information, while weighing costs and following steps to prepare data, train, deploy, and test.

Discover the fine tuning use case, data formats, and roles (system, user, assistant) for building jsonl training and validation sets, with at least ten example messages to tune models.

Create an Azure AI resource in the appropriate region, deploy the Azure OpenAI service, then access Azure AI Studio to customize the GPT 3.5 turbo model.

Explore fine tuning with Azure OpenAI studio and demonstrate Python API workflows to address current Java API limitations, while applying fine tuning from the studio to your programs.

Upload the training and validation json files in azure openai studio, confirm they are processed, then create a custom model from gpt-3.5 turbo and start the fine-tuning job.

Learn to fine tune a model using the OpenAI API with Python by uploading training and validation files, starting a fine tuning job (GPT-3.5-turbo-0613), and tracking status via IDs.

Apply fine tuning via the api with a deliberate delay between upload and training, monitor job status and deployment in the studio, and deploy the fine-tuned model.

Delete the deployment first to stop charges, then delete the fine-tuned model to complete cleanup.