Talk

AI technologies, and particularly large language models (LLMs), have been popping up like mushrooms lately. But how can you use them in your applications?

In this workshop, we will use a chatbot to interact with GPT-4 and implement the Retrieval Augmented Generation (RAG) pattern. Using a vector database, the model will be able to answer questions in natural language and generate complete, sourced responses from your own documents. To do this, we will create a Quarkus service based on the Open Source LangChain4J and ChatBootAI frameworks to test our chatbot. Finally, we will deploy everything to the Cloud.

After a short introduction to language models (operations and limitations), and prompt engineering, you will:
- Create a knowledge base: local HuggingFace LLMs, embeddings, a vector database, and semantic search
- Use LangChain4J to implement the RAG (Retrieval Augmented Generation) pattern
- Create a Quarkus API to interact with the LLM: OpenAI / AzureOpenAI
- Use ChatBootAI to interact with the Quarkus API
- Improve performance thanks to prompt engineering
- Containerize the application
- Deploy the containerized application to the Cloud
- Tweak your RAG integration
- Optimize for quality, cost or size

At the end of the workshop, you will have a clearer understanding of large language models and how they work, as well as ideas for using them in your applications. You will also know how to create a functional knowledge base and chatbot, and how to deploy them in the cloud.