Description
Generative AI creates numerous new opportunities for companies, especially when it comes to increasing productivity and efficiency. It makes features cheaper or even feasible in the first place. However, as these models are not deterministic, they occasionally provide incorrect or unusable information. In this workshop, we will shed light on the properties of generative models and understand the causes of these challenges. We will learn how to feed the models with our own data in order to improve the results. We will use the Retrieval Augmented Generation (RAG) architecture.
The aim of the course is to get to know all the building blocks of a simple RAG architecture. The experiments are designed to introduce these building blocks step by step. At the end of the workshop, we will combine all components into a chatbot that works with our own documents.
Agenda
Day 1
- Fundamentals of Large Language Models (LLMs)
- Architecture
- Model parameters
- Prompting
- Working with the OpenAI Chat API
- Building a simple LLM chatbot
- First experiments with the chatbot
Day 2
- Function Calling: The LLM as an agent
- Implementing your own functions for the LLM
- Basics of embeddings
- Semantic representation with vectors
- Determining text similarity
- Retrieval Augmented Generation (RAG)
- Use cases
- Preparing company data for LLMs
- Vector databases
- Make company data available for the chatbot
- Test LLMs
- When are generated answers “good”?
- Use embeddings to create test scenarios
- Use user feedback to improve the answers
Your Benefits
You will learn how to use the OpenAI Chat API.
You will understand how to supply an LLM with data.
You will learn how to integrate an LLM into existing systems.
You will be able to better assess how GenAI can be used in your environment.
Audience
Developers
Prerequisites
The workshop uses Jupyter Notebooks. The code to be written is simple. Basic knowledge of Python is sufficient, and an AI assistant will support you if required.
Technical requirements
- Python 3.10 or higher
- Jupyter
- C++ compiler (Visual Studio C++ Buildtools under Windows, under Linux or Mac the python-dev (apt) or python-devel (brew) package)
- Optional: Preferred IDE with Python and Jupyter plugin. Alternatively, JupyterLab offers everything you need.
Training Objectives
Using the OpenAI Chat API
Understanding prompting techniques
Recognize approaches against hallucinations of LLMs
Understanding the structure of an LLM chatbot
Making company data available with retrieval augmented generation
Extending LLMs with your own tools
Understanding embeddings and their application
Your Trainers
Hermann Schmidt
INNOQ
Facilitation, large language models
- GenAI for Developers
Hermann Schmidt works as a Senior Consultant at INNOQ. After more than two decades as a developer and architect, which mainly revolved around the “how” of software development, he is now focusing on the “what” and “who”. As a facilitator, he is interested in team structures, development and innovation processes, as well as creative techniques. Problems that hide in the cloud between the specialist department and the development team are his favorite area. Lately, Large Language Models have ignited a spark in him that reminds him of the time as a 17-year-old in high school when he sat wide-eyed in fascination in front of the only computer and wrote his first programs.
Marco Steinke
INNOQ
Software architecture, AI
- GenAI for Developers
Marco Steinke is a consultant at INNOQ. His focus is on software architecture. He also deals with artificial intelligence, particularly the architecture and integration of AI systems.
Technical Information and Books
Generative AI
The End of “Too Expensive” in Business Software? Exploring Features That Were Once Out of Reach. Zum Blogpost
Relevant Other Training Courses
by Request
Workshop
If you would like an entertaining introduction to prompt engineering.