Bring a crew to do your job

Everybody is talking about AGI, Artificial General Intelligence, or one AI Model that can do everything. Looking at real-world scenarios, there is not one person who can do everything. It takes teamwork to get things done. I see the exact solution for AI. You need a team, a crew, or a swarm to get things done.

Discover the power of multi-agent AI systems, where multiple agents collaborate to tackle challenges too complex for a single model. This talk will introduce core concepts of multi-agent systems, highlight some use cases, and feature a live demo showcasing agents working together in real-time. Learn how cooperative AI can create adaptive, resilient solutions to real-world problems.

Audience Takeaways: Attendees will gain a practical understanding of multi-agent AI systems, insights into real-world applications, and inspiration for how cooperative AI can address complex challenges in their work.

LLMs need a good retriever

LLMs are all around you. You cannot think of a conference that does not talk about large language models. In my line of business, I run into LLM applications regularly. Often, these encounters include RAG, short for Retrieval Augmented Generation. RAG solutions provide the knowledge that LLMs do not have themselves. Some examples are recent and private content. In short, for RAG, you create vectors from your content, store them in a vector store, fetch the best-matching chunks to a question, and return these as a context to a large language model that generates an answer to your question.
The retrieval aspect of RAG is more nuanced than it may seem. How do you determine the optimal chunks of content for a vector? Is a sentence, max tokens, or something else the best approach? Are these chunks indeed the most suitable context for the LLM? How can you ensure that your results perfectly fit the posed question?
If these questions sound familiar or if you want to learn more about RAG systems, this talk is for you. For the demos, I use my framework, "RAG4j," to interact with different LLMs and to create embeddings. The retriever is OpenSearch.

I do not use slides during this talk. It is a pure demo presentation. In the demo, I discuss the different aspects of RAG through examples.

Build the best knowledge retriever for your Large Language Model.

Generative AI is here to stay. Tools to generate text, images, or data are now common goods. Large Language models (LLMs) only have the knowledge they acquired through learning, and even that knowledge does not include all the details. To overcome the knowledge problem, the Retrieval Augmented Generation (RAG) pattern arose. An essential part of RAG is the retrieval part. Retrieval is not new. The search or retrieval domain is rich with tools, metrics and research. The new kid on the block is semantic search using vectors. Vector search got a jump start with the rise of LLMs and RAG.
This workshop aims to build a high-quality retriever, integrate the retriever into your LLM solution and measure the overall quality of your RAG system.
The workshop uses our Rag4j/Rag4p framework, which we created especially for workshops. It is easy to learn, so you can focus on understanding and building the details of the components during the workshop. You experiment with different chunking mechanisms (sentence, max tokens, semantic). After that, you use various strategies to construct the context for the LLM (TopN, Window, Document, Hierarchical). To find the optimum combination, you'll use quality metrics for the retriever as well as the other components of the RAG system.
You can do the workshop using Python or Java. We provide access to a remote LLM. You can also run an open-source LLM on Ollama on your local machine.

The Art of Questions: Creating a Semantic Search-Based Question-Answering System with LLMs

Ever thought about building your very own question-answering system? Like the one that powers Siri, Alexa, or Google Assistant? Well, we've got something awesome lined up for you!

In our hands-on workshop, we'll guide you through the ins and outs of creating a question-answering system. We prefer using Python for the workshop. We have prepared a GUI that works with python. If you prefer another language, you can still do the workshop, but you will miss the GUI to test your application. You'll get your hands dirty with vector stores and Large Language Models, we help you combine these two in a way you've never done before.

You've probably used search engines for keyword-based searches, right? Well, prepare to have your mind blown. We'll dive into something called semantic search, which is the next big thing after traditional searches. It’s like moving from asking Google to search "best pizza places" to "Where can I find a pizza place that my gluten-intolerant, vegan friend would love?" – you get the idea, right?

We’ll be teaching you how to build an entire pipeline, starting from collecting data from various sources, converting that into vectors (yeah, it’s more math, but it’s cool, we promise), and storing it so you can use it to answer all sorts of queries. It's like building your own mini Google!

We've got a repository ready to help you set up everything you need on your laptop. By the end of our workshop, you'll have your question-answering system ready and running.

So, why wait? Grab your laptop, bring your coding hat, and let's start building something fantastic together. Trust us, it’s going to be a blast!

Some of the highlights of the workshop:

Use a vector store (OpenSearch, Elasticsearch, Weaviate)
Use a Large Language Model (OpenAI, HuggingFace, Cohere, PaLM, Bedrock)
Use a tool for content extraction (Unstructured, Llama)
Create your pipeline (Langchain, Custom)

This is a workshop

EventStorming, from Knowledge to Working Domain Model

Two brains, one domain, one laptop, and a guarantee for a day of fun and learning. You’ll go back to basics. Talking about the domain, exploring it using EventStorming with a small group and then pair up to design and code the model.

## Agenda:
The day starts with an EventStorming session to explore an interesting domain. You will work in small groups, making sure everyone in the group guards a portion of the EventStorming session by taking on specific roles. But we won't stop after we've explored the domain.
During the second part of the day you will work in pairs, together with other participants. You will use the output of the EventStorming session and all the knowledge that you've gathered about the domain. Together, you'll play around, modeling parts of the domain in code, using tests to guide your design. You can choose from several starter templates, to help you code the domain using different approaches. From a typical object oriented domain model to an event sourced domain model. Throughout the workshop, pairs will share their approach so we can all learn from each other.

All you have to bring is a laptop with a recent version of IntelliJ installed (only one laptop per pair is needed). Java or Kotlin templates will be provided, or you can choose to use your own preferred stack.

## What you will learn:
- Why you need to start by understanding the problem and exploring a domain before you start coding.
- How to 'sketch' in code and tease out a model bit by bit.
- Not to be afraid to throw away the model and sketch again.
- That there are many ways to implement a model and that there are tradeoffs to make when choosing an approach.

Especially the part where pairs share and explain their approach is really insightful and helps to realize that there are many (good) solutions to a problem. It also creates a lot of interaction between participants which always makes for many views being shared and for lively discussions.

Playing with domain models in code

(This is a hands-on lab with limited capacity)

Two brains, one domain, one laptop, and a guarantee for 2 hours of fun and learning. You’ll go back to basics. Talking about the domain, designing the model and explore coding the model in Java or Kotlin with close to no libraries.

This is a hands-on coding workshop where you will work in pairs together with other participants. Your base will be the output of an EventStorming session and some context about the domain that you'll be modeling. Together, you'll play around, modeling parts of the domain in code, using tests to guide your design.

A starter project will be provided to get you up and running quickly, you can choose between Java or Kotlin. All you have to bring is a laptop with a recent version of IntelliJ installed (only one laptop per pair is needed). Throughout the workshop, pairs will share their approach so we can all learn from each other.

What you will learn:

- Why you need to start by understanding the problem before you start coding.
- How to 'sketch' in code and tease out a model bit by bit.
- Not to be afraid to throw away the model and sketch again.

The Good, the Bad and the Ugly data

In the cinematic masterpiece "The Good, the Bad, and the Ugly," characters mirror the nuances of data in analysis. "The Good" stands for high-quality, reliable data with a clear lineage; "The Bad" signifies anomalies in need of valorization efforts; and "The Ugly" represents unstructured data with hidden potential.

This presentation explores strategies for identifying and extracting the gold nuggets of insight from the "Ugly" data while mitigating the influence of the "Bad" data. We show data cleaning, validation, transformation, and presentation techniques to turn ugly data into good data, remove bad data and extract value from the good data.